lustre/lfsck/lfsck_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2013, 2016, Intel Corporation.
  24  */
  25 /*
  26  * lustre/lfsck/lfsck_lib.c
  27  *
  28  * Author: Fan, Yong <fan.yong@intel.com>
  29  */
  30
  31 #define DEBUG_SUBSYSTEM S_LFSCK
  32
  33 #include <linux/kthread.h>
  34 #include <linux/sched.h>
  35 #include <linux/list.h>
  36 #include <lu_object.h>
  37 #include <dt_object.h>
  38 #include <md_object.h>
  39 #include <lustre_fld.h>
  40 #include <lustre_lib.h>
  41 #include <lustre_net.h>
  42 #include <lustre_lfsck.h>
  43
  44 #include "lfsck_internal.h"
  45
  46 #define LFSCK_CHECKPOINT_SKIP   1
  47
  48 /* define lfsck thread key */
  49 LU_KEY_INIT(lfsck, struct lfsck_thread_info);
  50
  51 static void lfsck_key_fini(const struct lu_context *ctx,
  52                            struct lu_context_key *key, void *data)
  53 {
  54         struct lfsck_thread_info *info = data;
  55
  56         lu_buf_free(&info->lti_linkea_buf);
  57         lu_buf_free(&info->lti_linkea_buf2);
  58         lu_buf_free(&info->lti_big_buf);
  59         OBD_FREE_PTR(info);
  60 }
  61
  62 LU_CONTEXT_KEY_DEFINE(lfsck, LCT_MD_THREAD | LCT_DT_THREAD);
  63 LU_KEY_INIT_GENERIC(lfsck);
  64
  65 static struct list_head lfsck_instance_list;
  66 static struct list_head lfsck_ost_orphan_list;
  67 static struct list_head lfsck_mdt_orphan_list;
  68 static DEFINE_SPINLOCK(lfsck_instance_lock);
  69
  70 const char *lfsck_flags_names[] = {
  71         "scanned-once",
  72         "inconsistent",
  73         "upgrade",
  74         "incomplete",
  75         "crashed_lastid",
  76         NULL
  77 };
  78
  79 const char *lfsck_param_names[] = {
  80         NULL,
  81         "failout",
  82         "dryrun",
  83         "all_targets",
  84         "broadcast",
  85         "orphan",
  86         "create_ostobj",
  87         "create_mdtobj",
  88         NULL,
  89         "delay_create_ostobj",
  90         NULL
  91 };
  92
  93 enum lfsck_verify_lpf_types {
  94         LVLT_BY_BOOKMARK        = 0,
  95         LVLT_BY_NAMEENTRY       = 1,
  96 };
  97
  98 static inline void
  99 lfsck_reset_ltd_status(struct lfsck_tgt_desc *ltd, enum lfsck_type type)
 100 {
 101         if (type == LFSCK_TYPE_LAYOUT) {
 102                 ltd->ltd_layout_status = LS_MAX;
 103                 ltd->ltd_layout_repaired = 0;
 104         } else {
 105                 ltd->ltd_namespace_status = LS_MAX;
 106                 ltd->ltd_namespace_repaired = 0;
 107         }
 108 }
 109
 110 static int lfsck_tgt_descs_init(struct lfsck_tgt_descs *ltds)
 111 {
 112         spin_lock_init(&ltds->ltd_lock);
 113         init_rwsem(&ltds->ltd_rw_sem);
 114         INIT_LIST_HEAD(&ltds->ltd_orphan);
 115         ltds->ltd_tgts_bitmap = CFS_ALLOCATE_BITMAP(BITS_PER_LONG);
 116         if (ltds->ltd_tgts_bitmap == NULL)
 117                 return -ENOMEM;
 118
 119         return 0;
 120 }
 121
 122 static void lfsck_tgt_descs_fini(struct lfsck_tgt_descs *ltds)
 123 {
 124         struct lfsck_tgt_desc   *ltd;
 125         struct lfsck_tgt_desc   *next;
 126         int                      idx;
 127
 128         down_write(&ltds->ltd_rw_sem);
 129
 130         list_for_each_entry_safe(ltd, next, &ltds->ltd_orphan,
 131                                  ltd_orphan_list) {
 132                 list_del_init(&ltd->ltd_orphan_list);
 133                 lfsck_tgt_put(ltd);
 134         }
 135
 136         if (unlikely(ltds->ltd_tgts_bitmap == NULL)) {
 137                 up_write(&ltds->ltd_rw_sem);
 138
 139                 return;
 140         }
 141
 142         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
 143                 ltd = lfsck_ltd2tgt(ltds, idx);
 144                 if (likely(ltd != NULL)) {
 145                         LASSERT(list_empty(&ltd->ltd_layout_list));
 146                         LASSERT(list_empty(&ltd->ltd_layout_phase_list));
 147                         LASSERT(list_empty(&ltd->ltd_namespace_list));
 148                         LASSERT(list_empty(&ltd->ltd_namespace_phase_list));
 149
 150                         ltds->ltd_tgtnr--;
 151                         cfs_bitmap_clear(ltds->ltd_tgts_bitmap, idx);
 152                         lfsck_assign_tgt(ltds, NULL, idx);
 153                         lfsck_tgt_put(ltd);
 154                 }
 155         }
 156
 157         LASSERTF(ltds->ltd_tgtnr == 0, "tgt count unmatched: %d\n",
 158                  ltds->ltd_tgtnr);
 159
 160         for (idx = 0; idx < TGT_PTRS; idx++) {
 161                 if (ltds->ltd_tgts_idx[idx] != NULL) {
 162                         OBD_FREE_PTR(ltds->ltd_tgts_idx[idx]);
 163                         ltds->ltd_tgts_idx[idx] = NULL;
 164                 }
 165         }
 166
 167         CFS_FREE_BITMAP(ltds->ltd_tgts_bitmap);
 168         ltds->ltd_tgts_bitmap = NULL;
 169         up_write(&ltds->ltd_rw_sem);
 170 }
 171
 172 static int __lfsck_add_target(const struct lu_env *env,
 173                               struct lfsck_instance *lfsck,
 174                               struct lfsck_tgt_desc *ltd,
 175                               bool for_ost, bool locked)
 176 {
 177         struct lfsck_tgt_descs *ltds;
 178         __u32                   index = ltd->ltd_index;
 179         int                     rc    = 0;
 180         ENTRY;
 181
 182         if (for_ost)
 183                 ltds = &lfsck->li_ost_descs;
 184         else
 185                 ltds = &lfsck->li_mdt_descs;
 186
 187         if (!locked)
 188                 down_write(&ltds->ltd_rw_sem);
 189
 190         LASSERT(ltds->ltd_tgts_bitmap != NULL);
 191
 192         if (index >= ltds->ltd_tgts_bitmap->size) {
 193                 __u32 newsize = max((__u32)ltds->ltd_tgts_bitmap->size,
 194                                     (__u32)BITS_PER_LONG);
 195                 struct cfs_bitmap *old_bitmap = ltds->ltd_tgts_bitmap;
 196                 struct cfs_bitmap *new_bitmap;
 197
 198                 while (newsize < index + 1)
 199                         newsize <<= 1;
 200
 201                 new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
 202                 if (new_bitmap == NULL)
 203                         GOTO(unlock, rc = -ENOMEM);
 204
 205                 if (ltds->ltd_tgtnr > 0)
 206                         cfs_bitmap_copy(new_bitmap, old_bitmap);
 207                 ltds->ltd_tgts_bitmap = new_bitmap;
 208                 CFS_FREE_BITMAP(old_bitmap);
 209         }
 210
 211         if (cfs_bitmap_check(ltds->ltd_tgts_bitmap, index)) {
 212                 CERROR("%s: the device %s (%u) is registered already\n",
 213                        lfsck_lfsck2name(lfsck),
 214                        ltd->ltd_tgt->dd_lu_dev.ld_obd->obd_name, index);
 215                 GOTO(unlock, rc = -EEXIST);
 216         }
 217
 218         if (ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 219                 OBD_ALLOC_PTR(ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK]);
 220                 if (ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK] == NULL)
 221                         GOTO(unlock, rc = -ENOMEM);
 222         }
 223
 224         lfsck_assign_tgt(ltds, ltd, index);
 225         cfs_bitmap_set(ltds->ltd_tgts_bitmap, index);
 226         ltds->ltd_tgtnr++;
 227
 228         GOTO(unlock, rc = 0);
 229
 230 unlock:
 231         if (!locked)
 232                 up_write(&ltds->ltd_rw_sem);
 233
 234         return rc;
 235 }
 236
 237 static int lfsck_add_target_from_orphan(const struct lu_env *env,
 238                                         struct lfsck_instance *lfsck)
 239 {
 240         struct lfsck_tgt_descs  *ltds    = &lfsck->li_ost_descs;
 241         struct lfsck_tgt_desc   *ltd;
 242         struct lfsck_tgt_desc   *next;
 243         struct list_head        *head    = &lfsck_ost_orphan_list;
 244         int                      rc;
 245         bool                     for_ost = true;
 246
 247 again:
 248         spin_lock(&lfsck_instance_lock);
 249         list_for_each_entry_safe(ltd, next, head, ltd_orphan_list) {
 250                 if (ltd->ltd_key == lfsck->li_bottom)
 251                         list_move_tail(&ltd->ltd_orphan_list,
 252                                        &ltds->ltd_orphan);
 253         }
 254         spin_unlock(&lfsck_instance_lock);
 255
 256         down_write(&ltds->ltd_rw_sem);
 257         while (!list_empty(&ltds->ltd_orphan)) {
 258                 ltd = list_entry(ltds->ltd_orphan.next,
 259                                  struct lfsck_tgt_desc,
 260                                  ltd_orphan_list);
 261                 list_del_init(&ltd->ltd_orphan_list);
 262                 rc = __lfsck_add_target(env, lfsck, ltd, for_ost, true);
 263                 /* Do not hold the semaphore for too long time. */
 264                 up_write(&ltds->ltd_rw_sem);
 265                 if (rc != 0)
 266                         return rc;
 267
 268                 down_write(&ltds->ltd_rw_sem);
 269         }
 270         up_write(&ltds->ltd_rw_sem);
 271
 272         if (for_ost) {
 273                 ltds = &lfsck->li_mdt_descs;
 274                 head = &lfsck_mdt_orphan_list;
 275                 for_ost = false;
 276                 goto again;
 277         }
 278
 279         return 0;
 280 }
 281
 282 static inline struct lfsck_component *
 283 __lfsck_component_find(struct lfsck_instance *lfsck, __u16 type,
 284                        struct list_head *list)
 285 {
 286         struct lfsck_component *com;
 287
 288         list_for_each_entry(com, list, lc_link) {
 289                 if (com->lc_type == type)
 290                         return com;
 291         }
 292         return NULL;
 293 }
 294
 295 struct lfsck_component *
 296 lfsck_component_find(struct lfsck_instance *lfsck, __u16 type)
 297 {
 298         struct lfsck_component *com;
 299
 300         spin_lock(&lfsck->li_lock);
 301         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_scan);
 302         if (com != NULL)
 303                 goto unlock;
 304
 305         com = __lfsck_component_find(lfsck, type,
 306                                      &lfsck->li_list_double_scan);
 307         if (com != NULL)
 308                 goto unlock;
 309
 310         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_idle);
 311
 312 unlock:
 313         if (com != NULL)
 314                 lfsck_component_get(com);
 315         spin_unlock(&lfsck->li_lock);
 316         return com;
 317 }
 318
 319 void lfsck_component_cleanup(const struct lu_env *env,
 320                              struct lfsck_component *com)
 321 {
 322         if (!list_empty(&com->lc_link))
 323                 list_del_init(&com->lc_link);
 324         if (!list_empty(&com->lc_link_dir))
 325                 list_del_init(&com->lc_link_dir);
 326
 327         lfsck_component_put(env, com);
 328 }
 329
 330 int lfsck_fid_alloc(const struct lu_env *env, struct lfsck_instance *lfsck,
 331                     struct lu_fid *fid, bool locked)
 332 {
 333         struct lfsck_bookmark   *bk = &lfsck->li_bookmark_ram;
 334         int                      rc = 0;
 335         ENTRY;
 336
 337         if (!locked)
 338                 mutex_lock(&lfsck->li_mutex);
 339
 340         rc = seq_client_alloc_fid(env, lfsck->li_seq, fid);
 341         if (rc >= 0) {
 342                 bk->lb_last_fid = *fid;
 343                 /* We do not care about whether the subsequent sub-operations
 344                  * failed or not. The worst case is that one FID is lost that
 345                  * is not a big issue for the LFSCK since it is relative rare
 346                  * for LFSCK create. */
 347                 rc = lfsck_bookmark_store(env, lfsck);
 348         }
 349
 350         if (!locked)
 351                 mutex_unlock(&lfsck->li_mutex);
 352
 353         RETURN(rc);
 354 }
 355
 356 static int __lfsck_ibits_lock(const struct lu_env *env,
 357                               struct lfsck_instance *lfsck,
 358                               struct dt_object *obj, struct ldlm_res_id *resid,
 359                               struct lustre_handle *lh, __u64 bits,
 360                               enum ldlm_mode mode)
 361 {
 362         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 363         union ldlm_policy_data          *policy = &info->lti_policy;
 364         __u64                            flags  = LDLM_FL_ATOMIC_CB;
 365         int                              rc;
 366
 367         LASSERT(lfsck->li_namespace != NULL);
 368
 369         memset(policy, 0, sizeof(*policy));
 370         policy->l_inodebits.bits = bits;
 371         if (dt_object_remote(obj)) {
 372                 struct ldlm_enqueue_info *einfo = &info->lti_einfo;
 373
 374                 memset(einfo, 0, sizeof(*einfo));
 375                 einfo->ei_type = LDLM_IBITS;
 376                 einfo->ei_mode = mode;
 377                 einfo->ei_cb_bl = ldlm_blocking_ast;
 378                 einfo->ei_cb_cp = ldlm_completion_ast;
 379                 einfo->ei_res_id = resid;
 380
 381                 rc = dt_object_lock(env, obj, lh, einfo, policy);
 382         } else {
 383                 rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid,
 384                                             LDLM_IBITS, policy, mode,
 385                                             &flags, ldlm_blocking_ast,
 386                                             ldlm_completion_ast, NULL, NULL,
 387                                             0, LVB_T_NONE, NULL, lh);
 388         }
 389
 390         if (rc == ELDLM_OK) {
 391                 rc = 0;
 392         } else {
 393                 memset(lh, 0, sizeof(*lh));
 394                 rc = -EIO;
 395         }
 396
 397         return rc;
 398 }
 399
 400 /**
 401  * Request the specified ibits lock for the given object.
 402  *
 403  * Before the LFSCK modifying on the namespace visible object,
 404  * it needs to acquire related ibits ldlm lock.
 405  *
 406  * \param[in] env       pointer to the thread context
 407  * \param[in] lfsck     pointer to the lfsck instance
 408  * \param[in] obj       pointer to the dt_object to be locked
 409  * \param[out] lh       pointer to the lock handle
 410  * \param[in] bits      the bits for the ldlm lock to be acquired
 411  * \param[in] mode      the mode for the ldlm lock to be acquired
 412  *
 413  * \retval              0 for success
 414  * \retval              negative error number on failure
 415  */
 416 int lfsck_ibits_lock(const struct lu_env *env, struct lfsck_instance *lfsck,
 417                      struct dt_object *obj, struct lustre_handle *lh,
 418                      __u64 bits, enum ldlm_mode mode)
 419 {
 420         struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid;
 421
 422         LASSERT(!lustre_handle_is_used(lh));
 423
 424         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
 425         return __lfsck_ibits_lock(env, lfsck, obj, resid, lh, bits, mode);
 426 }
 427
 428 /**
 429  * Release the the specified ibits lock.
 430  *
 431  * If the lock has been acquired before, release it
 432  * and cleanup the handle. Otherwise, do nothing.
 433  *
 434  * \param[in] lh        pointer to the lock handle
 435  * \param[in] mode      the mode for the ldlm lock to be released
 436  */
 437 void lfsck_ibits_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
 438 {
 439         if (lustre_handle_is_used(lh)) {
 440                 ldlm_lock_decref(lh, mode);
 441                 memset(lh, 0, sizeof(*lh));
 442         }
 443 }
 444
 445 /**
 446  * Request compound ibits locks for the given <obj, name> pairs.
 447  *
 448  * Before the LFSCK modifying on the namespace visible object, it needs to
 449  * acquire related ibits ldlm lock. Usually, we can use lfsck_ibits_lock for
 450  * the lock purpose. But the simple lfsck_ibits_lock for directory-based
 451  * modificationis (such as insert name entry to the directory) may be too
 452  * coarse-grained and not efficient.
 453  *
 454  * The lfsck_lock() will request compound ibits locks on the specified
 455  * <obj, name> pairs: the PDO (Parallel Directory Operations) ibits (UPDATE)
 456  * lock on the directory object, and the regular ibits lock on the name hash.
 457  *
 458  * \param[in] env       pointer to the thread context
 459  * \param[in] lfsck     pointer to the lfsck instance
 460  * \param[in] obj       pointer to the dt_object to be locked
 461  * \param[in] name      used for building the PDO lock resource
 462  * \param[out] llh      pointer to the lfsck_lock_handle
 463  * \param[in] bits      the bits for the ldlm lock to be acquired
 464  * \param[in] mode      the mode for the ldlm lock to be acquired
 465  *
 466  * \retval              0 for success
 467  * \retval              negative error number on failure
 468  */
 469 int lfsck_lock(const struct lu_env *env, struct lfsck_instance *lfsck,
 470                struct dt_object *obj, const char *name,
 471                struct lfsck_lock_handle *llh, __u64 bits, enum ldlm_mode mode)
 472 {
 473         struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid;
 474         int                 rc;
 475
 476         LASSERT(S_ISDIR(lfsck_object_type(obj)));
 477         LASSERT(name != NULL);
 478         LASSERT(name[0] != 0);
 479         LASSERT(!lustre_handle_is_used(&llh->llh_pdo_lh));
 480         LASSERT(!lustre_handle_is_used(&llh->llh_reg_lh));
 481
 482         switch (mode) {
 483         case LCK_EX:
 484                 llh->llh_pdo_mode = LCK_EX;
 485                 break;
 486         case LCK_PW:
 487                 llh->llh_pdo_mode = LCK_CW;
 488                 break;
 489         case LCK_PR:
 490                 llh->llh_pdo_mode = LCK_CR;
 491                 break;
 492         default:
 493                 CDEBUG(D_LFSCK, "%s: unexpected PDO lock mode %u on the obj "
 494                        DFID"\n", lfsck_lfsck2name(lfsck), mode,
 495                        PFID(lfsck_dto2fid(obj)));
 496                 LBUG();
 497         }
 498
 499         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
 500         rc = __lfsck_ibits_lock(env, lfsck, obj, resid, &llh->llh_pdo_lh,
 501                                 MDS_INODELOCK_UPDATE, llh->llh_pdo_mode);
 502         if (rc != 0)
 503                 return rc;
 504
 505         llh->llh_reg_mode = mode;
 506         resid->name[LUSTRE_RES_ID_HSH_OFF] = ll_full_name_hash(NULL, name,
 507                                                                strlen(name));
 508         LASSERT(resid->name[LUSTRE_RES_ID_HSH_OFF] != 0);
 509         rc = __lfsck_ibits_lock(env, lfsck, obj, resid, &llh->llh_reg_lh,
 510                                 bits, llh->llh_reg_mode);
 511         if (rc != 0)
 512                 lfsck_ibits_unlock(&llh->llh_pdo_lh, llh->llh_pdo_mode);
 513
 514         return rc;
 515 }
 516
 517 /**
 518  * Release the the compound ibits locks.
 519  *
 520  * \param[in] llh       pointer to the lfsck_lock_handle to be released
 521  */
 522 void lfsck_unlock(struct lfsck_lock_handle *llh)
 523 {
 524         lfsck_ibits_unlock(&llh->llh_reg_lh, llh->llh_reg_mode);
 525         lfsck_ibits_unlock(&llh->llh_pdo_lh, llh->llh_pdo_mode);
 526 }
 527
 528 int lfsck_find_mdt_idx_by_fid(const struct lu_env *env,
 529                               struct lfsck_instance *lfsck,
 530                               const struct lu_fid *fid)
 531 {
 532         struct seq_server_site  *ss     = lfsck_dev_site(lfsck);
 533         struct lu_seq_range     *range  = &lfsck_env_info(env)->lti_range;
 534         int                      rc;
 535
 536         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
 537                 /* "ROOT" is always on the MDT0. */
 538                 if (lu_fid_eq(fid, &lfsck->li_global_root_fid))
 539                         return 0;
 540
 541                 return lfsck_dev_idx(lfsck);
 542         }
 543
 544         fld_range_set_mdt(range);
 545         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
 546         if (rc == 0)
 547                 rc = range->lsr_index;
 548
 549         return rc;
 550 }
 551
 552 const char dot[] = ".";
 553 const char dotdot[] = "..";
 554 static const char dotlustre[] = ".lustre";
 555 static const char lostfound[] = "lost+found";
 556
 557 /**
 558  * Remove the name entry from the .lustre/lost+found directory.
 559  *
 560  * No need to care about the object referenced by the name entry,
 561  * either the name entry is invalid or redundant, or the referenced
 562  * object has been processed or will be handled by others.
 563  *
 564  * \param[in] env       pointer to the thread context
 565  * \param[in] lfsck     pointer to the lfsck instance
 566  * \param[in] name      the name for the name entry to be removed
 567  *
 568  * \retval              0 for success
 569  * \retval              negative error number on failure
 570  */
 571 static int lfsck_lpf_remove_name_entry(const struct lu_env *env,
 572                                        struct lfsck_instance *lfsck,
 573                                        const char *name)
 574 {
 575         struct dt_object        *parent = lfsck->li_lpf_root_obj;
 576         struct dt_device        *dev    = lfsck_obj2dev(parent);
 577         struct thandle          *th;
 578         struct lfsck_lock_handle *llh   = &lfsck_env_info(env)->lti_llh;
 579         int                      rc;
 580         ENTRY;
 581
 582         rc = lfsck_lock(env, lfsck, parent, name, llh,
 583                         MDS_INODELOCK_UPDATE, LCK_PW);
 584         if (rc != 0)
 585                 RETURN(rc);
 586
 587         th = dt_trans_create(env, dev);
 588         if (IS_ERR(th))
 589                 GOTO(unlock, rc = PTR_ERR(th));
 590
 591         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
 592         if (rc != 0)
 593                 GOTO(stop, rc);
 594
 595         rc = dt_declare_ref_del(env, parent, th);
 596         if (rc != 0)
 597                 GOTO(stop, rc);
 598
 599         rc = dt_trans_start_local(env, dev, th);
 600         if (rc != 0)
 601                 GOTO(stop, rc);
 602
 603         rc = dt_delete(env, parent, (const struct dt_key *)name, th);
 604         if (rc != 0)
 605                 GOTO(stop, rc);
 606
 607         dt_write_lock(env, parent, 0);
 608         rc = dt_ref_del(env, parent, th);
 609         dt_write_unlock(env, parent);
 610
 611         GOTO(stop, rc);
 612
 613 stop:
 614         dt_trans_stop(env, dev, th);
 615
 616 unlock:
 617         lfsck_unlock(llh);
 618
 619         CDEBUG(D_LFSCK, "%s: remove name entry "DFID"/%s: rc = %d\n",
 620                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(parent)), name, rc);
 621
 622         return rc;
 623 }
 624
 625 static int lfsck_create_lpf_local(const struct lu_env *env,
 626                                   struct lfsck_instance *lfsck,
 627                                   struct dt_object *child,
 628                                   struct lu_attr *la,
 629                                   struct dt_object_format *dof,
 630                                   const char *name)
 631 {
 632         struct dt_insert_rec    *rec    = &lfsck_env_info(env)->lti_dt_rec;
 633         struct dt_object        *parent = lfsck->li_lpf_root_obj;
 634         struct dt_device        *dev    = lfsck_obj2dev(child);
 635         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
 636         struct dt_object        *bk_obj = lfsck->li_bookmark_obj;
 637         const struct lu_fid     *cfid   = lfsck_dto2fid(child);
 638         struct thandle          *th     = NULL;
 639         struct linkea_data       ldata  = { NULL };
 640         struct lu_buf            linkea_buf;
 641         const struct lu_name    *cname;
 642         loff_t                   pos    = 0;
 643         int                      len    = sizeof(struct lfsck_bookmark);
 644         int                      rc;
 645         ENTRY;
 646
 647         cname = lfsck_name_get_const(env, name, strlen(name));
 648         rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf2,
 649                               cname, lfsck_dto2fid(parent));
 650         if (rc != 0)
 651                 RETURN(rc);
 652
 653         th = dt_trans_create(env, dev);
 654         if (IS_ERR(th))
 655                 RETURN(PTR_ERR(th));
 656
 657         /* 1a. create child */
 658         rc = dt_declare_create(env, child, la, NULL, dof, th);
 659         if (rc != 0)
 660                 GOTO(stop, rc);
 661
 662         if (!dt_try_as_dir(env, child))
 663                 GOTO(stop, rc = -ENOTDIR);
 664
 665         /* 2a. increase child nlink */
 666         rc = dt_declare_ref_add(env, child, th);
 667         if (rc != 0)
 668                 GOTO(stop, rc);
 669
 670         /* 3a. insert dot into child dir */
 671         rec->rec_type = S_IFDIR;
 672         rec->rec_fid = cfid;
 673         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 674                                (const struct dt_key *)dot, th);
 675         if (rc != 0)
 676                 GOTO(stop, rc);
 677
 678         /* 4a. insert dotdot into child dir */
 679         rec->rec_fid = &LU_LPF_FID;
 680         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 681                                (const struct dt_key *)dotdot, th);
 682         if (rc != 0)
 683                 GOTO(stop, rc);
 684
 685         /* 5a. insert linkEA for child */
 686         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
 687                        ldata.ld_leh->leh_len);
 688         rc = dt_declare_xattr_set(env, child, &linkea_buf,
 689                                   XATTR_NAME_LINK, 0, th);
 690         if (rc != 0)
 691                 GOTO(stop, rc);
 692
 693         /* 6a. insert name into parent dir */
 694         rec->rec_type = S_IFDIR;
 695         rec->rec_fid = cfid;
 696         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
 697                                (const struct dt_key *)name, th);
 698         if (rc != 0)
 699                 GOTO(stop, rc);
 700
 701         /* 7a. increase parent nlink */
 702         rc = dt_declare_ref_add(env, parent, th);
 703         if (rc != 0)
 704                 GOTO(stop, rc);
 705
 706         /* 8a. update bookmark */
 707         rc = dt_declare_record_write(env, bk_obj,
 708                                      lfsck_buf_get(env, bk, len), 0, th);
 709         if (rc != 0)
 710                 GOTO(stop, rc);
 711
 712         rc = dt_trans_start_local(env, dev, th);
 713         if (rc != 0)
 714                 GOTO(stop, rc);
 715
 716         dt_write_lock(env, child, 0);
 717         /* 1b. create child */
 718         rc = dt_create(env, child, la, NULL, dof, th);
 719         if (rc != 0)
 720                 GOTO(unlock, rc);
 721
 722         /* 2b. increase child nlink */
 723         rc = dt_ref_add(env, child, th);
 724         if (rc != 0)
 725                 GOTO(unlock, rc);
 726
 727         /* 3b. insert dot into child dir */
 728         rec->rec_fid = cfid;
 729         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 730                        (const struct dt_key *)dot, th, 1);
 731         if (rc != 0)
 732                 GOTO(unlock, rc);
 733
 734         /* 4b. insert dotdot into child dir */
 735         rec->rec_fid = &LU_LPF_FID;
 736         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 737                        (const struct dt_key *)dotdot, th, 1);
 738         if (rc != 0)
 739                 GOTO(unlock, rc);
 740
 741         /* 5b. insert linkEA for child. */
 742         rc = dt_xattr_set(env, child, &linkea_buf,
 743                           XATTR_NAME_LINK, 0, th);
 744         dt_write_unlock(env, child);
 745         if (rc != 0)
 746                 GOTO(stop, rc);
 747
 748         /* 6b. insert name into parent dir */
 749         rec->rec_fid = cfid;
 750         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
 751                        (const struct dt_key *)name, th, 1);
 752         if (rc != 0)
 753                 GOTO(stop, rc);
 754
 755         dt_write_lock(env, parent, 0);
 756         /* 7b. increase parent nlink */
 757         rc = dt_ref_add(env, parent, th);
 758         dt_write_unlock(env, parent);
 759         if (rc != 0)
 760                 GOTO(stop, rc);
 761
 762         bk->lb_lpf_fid = *cfid;
 763         lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
 764
 765         /* 8b. update bookmark */
 766         rc = dt_record_write(env, bk_obj,
 767                              lfsck_buf_get(env, bk, len), &pos, th);
 768
 769         GOTO(stop, rc);
 770
 771 unlock:
 772         dt_write_unlock(env, child);
 773
 774 stop:
 775         dt_trans_stop(env, dev, th);
 776
 777         return rc;
 778 }
 779
 780 static int lfsck_create_lpf_remote(const struct lu_env *env,
 781                                    struct lfsck_instance *lfsck,
 782                                    struct dt_object *child,
 783                                    struct lu_attr *la,
 784                                    struct dt_object_format *dof,
 785                                    const char *name)
 786 {
 787         struct dt_insert_rec    *rec    = &lfsck_env_info(env)->lti_dt_rec;
 788         struct dt_object        *parent = lfsck->li_lpf_root_obj;
 789         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
 790         struct dt_object        *bk_obj = lfsck->li_bookmark_obj;
 791         const struct lu_fid     *cfid   = lfsck_dto2fid(child);
 792         struct thandle          *th     = NULL;
 793         struct linkea_data       ldata  = { NULL };
 794         struct lu_buf            linkea_buf;
 795         const struct lu_name    *cname;
 796         struct dt_device        *dev;
 797         loff_t                   pos    = 0;
 798         int                      len    = sizeof(struct lfsck_bookmark);
 799         int                      rc;
 800         ENTRY;
 801
 802         cname = lfsck_name_get_const(env, name, strlen(name));
 803         rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf2,
 804                               cname, lfsck_dto2fid(parent));
 805         if (rc != 0)
 806                 RETURN(rc);
 807
 808         /* Create .lustre/lost+found/MDTxxxx. */
 809
 810         /* XXX: Currently, cross-MDT create operation needs to create the child
 811          *      object firstly, then insert name into the parent directory. For
 812          *      this case, the child object resides on current MDT (local), but
 813          *      the parent ".lustre/lost+found" may be on remote MDT. It is not
 814          *      easy to contain all the sub-modifications orderly within single
 815          *      transaction.
 816          *
 817          *      To avoid more inconsistency, we split the create operation into
 818          *      two transactions:
 819          *
 820          *      1) create the child and update the lfsck_bookmark::lb_lpf_fid
 821          *         locally.
 822          *      2) insert the name "MDTXXXX" in the parent ".lustre/lost+found"
 823          *         remotely.
 824          *
 825          *      If 1) done, but 2) failed, then go ahead, the LFSCK will try to
 826          *      repair such inconsistency when LFSCK run next time. */
 827
 828         /* Transaction I: locally */
 829
 830         dev = lfsck_obj2dev(child);
 831         th = dt_trans_create(env, dev);
 832         if (IS_ERR(th))
 833                 RETURN(PTR_ERR(th));
 834
 835         /* 1a. create child */
 836         rc = dt_declare_create(env, child, la, NULL, dof, th);
 837         if (rc != 0)
 838                 GOTO(stop, rc);
 839
 840         if (!dt_try_as_dir(env, child))
 841                 GOTO(stop, rc = -ENOTDIR);
 842
 843         /* 2a. increase child nlink */
 844         rc = dt_declare_ref_add(env, child, th);
 845         if (rc != 0)
 846                 GOTO(stop, rc);
 847
 848         /* 3a. insert dot into child dir */
 849         rec->rec_type = S_IFDIR;
 850         rec->rec_fid = cfid;
 851         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 852                                (const struct dt_key *)dot, th);
 853         if (rc != 0)
 854                 GOTO(stop, rc);
 855
 856         /* 4a. insert dotdot into child dir */
 857         rec->rec_fid = &LU_LPF_FID;
 858         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 859                                (const struct dt_key *)dotdot, th);
 860         if (rc != 0)
 861                 GOTO(stop, rc);
 862
 863         /* 5a. insert linkEA for child */
 864         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
 865                        ldata.ld_leh->leh_len);
 866         rc = dt_declare_xattr_set(env, child, &linkea_buf,
 867                                   XATTR_NAME_LINK, 0, th);
 868         if (rc != 0)
 869                 GOTO(stop, rc);
 870
 871         /* 6a. update bookmark */
 872         rc = dt_declare_record_write(env, bk_obj,
 873                                      lfsck_buf_get(env, bk, len), 0, th);
 874         if (rc != 0)
 875                 GOTO(stop, rc);
 876
 877         rc = dt_trans_start_local(env, dev, th);
 878         if (rc != 0)
 879                 GOTO(stop, rc);
 880
 881         dt_write_lock(env, child, 0);
 882         /* 1b. create child */
 883         rc = dt_create(env, child, la, NULL, dof, th);
 884         if (rc != 0)
 885                 GOTO(unlock, rc);
 886
 887         /* 2b. increase child nlink */
 888         rc = dt_ref_add(env, child, th);
 889         if (rc != 0)
 890                 GOTO(unlock, rc);
 891
 892         /* 3b. insert dot into child dir */
 893         rec->rec_type = S_IFDIR;
 894         rec->rec_fid = cfid;
 895         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 896                        (const struct dt_key *)dot, th, 1);
 897         if (rc != 0)
 898                 GOTO(unlock, rc);
 899
 900         /* 4b. insert dotdot into child dir */
 901         rec->rec_fid = &LU_LPF_FID;
 902         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 903                        (const struct dt_key *)dotdot, th, 1);
 904         if (rc != 0)
 905                 GOTO(unlock, rc);
 906
 907         /* 5b. insert linkEA for child */
 908         rc = dt_xattr_set(env, child, &linkea_buf,
 909                           XATTR_NAME_LINK, 0, th);
 910         if (rc != 0)
 911                 GOTO(unlock, rc);
 912
 913         bk->lb_lpf_fid = *cfid;
 914         lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
 915
 916         /* 6b. update bookmark */
 917         rc = dt_record_write(env, bk_obj,
 918                              lfsck_buf_get(env, bk, len), &pos, th);
 919
 920         dt_write_unlock(env, child);
 921         dt_trans_stop(env, dev, th);
 922         if (rc != 0)
 923                 RETURN(rc);
 924
 925         /* Transaction II: remotely */
 926
 927         dev = lfsck_obj2dev(parent);
 928         th = dt_trans_create(env, dev);
 929         if (IS_ERR(th))
 930                 RETURN(PTR_ERR(th));
 931
 932         th->th_sync = 1;
 933         /* 5a. insert name into parent dir */
 934         rec->rec_fid = cfid;
 935         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
 936                                (const struct dt_key *)name, th);
 937         if (rc != 0)
 938                 GOTO(stop, rc);
 939
 940         /* 6a. increase parent nlink */
 941         rc = dt_declare_ref_add(env, parent, th);
 942         if (rc != 0)
 943                 GOTO(stop, rc);
 944
 945         rc = dt_trans_start_local(env, dev, th);
 946         if (rc != 0)
 947                 GOTO(stop, rc);
 948
 949         /* 5b. insert name into parent dir */
 950         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
 951                        (const struct dt_key *)name, th, 1);
 952         if (rc != 0)
 953                 GOTO(stop, rc);
 954
 955         dt_write_lock(env, parent, 0);
 956         /* 6b. increase parent nlink */
 957         rc = dt_ref_add(env, parent, th);
 958         dt_write_unlock(env, parent);
 959
 960         GOTO(stop, rc);
 961
 962 unlock:
 963         dt_write_unlock(env, child);
 964 stop:
 965         dt_trans_stop(env, dev, th);
 966
 967         if (rc != 0 && dev == lfsck_obj2dev(parent))
 968                 CDEBUG(D_LFSCK, "%s: partially created the object "DFID
 969                        "for orphans, but failed to insert the name %s "
 970                        "to the .lustre/lost+found/. Such inconsistency "
 971                        "will be repaired when LFSCK run next time: rc = %d\n",
 972                        lfsck_lfsck2name(lfsck), PFID(cfid), name, rc);
 973
 974         return rc;
 975 }
 976
 977 /**
 978  * Create the MDTxxxx directory under /ROOT/.lustre/lost+found/
 979  *
 980  * The /ROOT/.lustre/lost+found/MDTxxxx/ directory is used for holding
 981  * orphans and other uncertain inconsistent objects found during the
 982  * LFSCK. Such directory will be created by the LFSCK engine on the
 983  * local MDT before the LFSCK scanning.
 984  *
 985  * \param[in] env       pointer to the thread context
 986  * \param[in] lfsck     pointer to the lfsck instance
 987  *
 988  * \retval              0 for success
 989  * \retval              negative error number on failure
 990  */
 991 static int lfsck_create_lpf(const struct lu_env *env,
 992                             struct lfsck_instance *lfsck)
 993 {
 994         struct lfsck_bookmark    *bk    = &lfsck->li_bookmark_ram;
 995         struct lfsck_thread_info *info  = lfsck_env_info(env);
 996         struct lu_fid            *cfid  = &info->lti_fid2;
 997         struct lu_attr           *la    = &info->lti_la;
 998         struct dt_object_format  *dof   = &info->lti_dof;
 999         struct dt_object         *parent = lfsck->li_lpf_root_obj;
1000         struct dt_object         *child = NULL;
1001         struct lfsck_lock_handle *llh   = &info->lti_llh;
1002         char                      name[8];
1003         int                       node  = lfsck_dev_idx(lfsck);
1004         int                       rc    = 0;
1005         ENTRY;
1006
1007         LASSERT(lfsck->li_master);
1008         LASSERT(parent != NULL);
1009         LASSERT(lfsck->li_lpf_obj == NULL);
1010
1011         snprintf(name, 8, "MDT%04x", node);
1012         rc = lfsck_lock(env, lfsck, parent, name, llh,
1013                         MDS_INODELOCK_UPDATE, LCK_PW);
1014         if (rc != 0)
1015                 RETURN(rc);
1016
1017         if (fid_is_zero(&bk->lb_lpf_fid)) {
1018                 /* There is corner case that: in former LFSCK scanning we have
1019                  * created the .lustre/lost+found/MDTxxxx but failed to update
1020                  * the lfsck_bookmark::lb_lpf_fid successfully. So need lookup
1021                  * it from MDT0 firstly. */
1022                 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1023                                (const struct dt_key *)name);
1024                 if (rc != 0 && rc != -ENOENT)
1025                         GOTO(unlock, rc);
1026
1027                 if (rc == 0) {
1028                         bk->lb_lpf_fid = *cfid;
1029                         rc = lfsck_bookmark_store(env, lfsck);
1030                 } else {
1031                         rc = lfsck_fid_alloc(env, lfsck, cfid, true);
1032                 }
1033                 if (rc != 0)
1034                         GOTO(unlock, rc);
1035         } else {
1036                 *cfid = bk->lb_lpf_fid;
1037         }
1038
1039         child = lfsck_object_find_bottom_new(env, lfsck, cfid);
1040         if (IS_ERR(child))
1041                 GOTO(unlock, rc = PTR_ERR(child));
1042
1043         if (dt_object_exists(child) != 0) {
1044                 if (unlikely(!dt_try_as_dir(env, child)))
1045                         rc = -ENOTDIR;
1046                 else
1047                         lfsck->li_lpf_obj = child;
1048
1049                 GOTO(unlock, rc);
1050         }
1051
1052         memset(la, 0, sizeof(*la));
1053         la->la_atime = la->la_mtime = la->la_ctime = ktime_get_real_seconds();
1054         la->la_mode = S_IFDIR | S_IRWXU;
1055         la->la_valid = LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1056                        LA_UID | LA_GID;
1057         memset(dof, 0, sizeof(*dof));
1058         dof->dof_type = dt_mode_to_dft(S_IFDIR);
1059
1060         if (node == 0)
1061                 rc = lfsck_create_lpf_local(env, lfsck, child, la, dof, name);
1062         else
1063                 rc = lfsck_create_lpf_remote(env, lfsck, child, la, dof, name);
1064         if (rc == 0)
1065                 lfsck->li_lpf_obj = child;
1066
1067         GOTO(unlock, rc);
1068
1069 unlock:
1070         lfsck_unlock(llh);
1071         if (rc != 0 && child != NULL && !IS_ERR(child))
1072                 lfsck_object_put(env, child);
1073
1074         return rc;
1075 }
1076
1077 /**
1078  * Scan .lustre/lost+found for bad name entries and remove them.
1079  *
1080  * The valid name entry should be "MDTxxxx", the "xxxx" is the MDT device
1081  * index in the system. Any other formatted name is invalid and should be
1082  * removed.
1083  *
1084  * \param[in] env       pointer to the thread context
1085  * \param[in] lfsck     pointer to the lfsck instance
1086  *
1087  * \retval              0 for success
1088  * \retval              negative error number on failure
1089  */
1090 static int lfsck_scan_lpf_bad_entries(const struct lu_env *env,
1091                                       struct lfsck_instance *lfsck)
1092 {
1093         struct dt_object        *parent = lfsck->li_lpf_root_obj;
1094         struct lu_dirent        *ent    =
1095                         (struct lu_dirent *)lfsck_env_info(env)->lti_key;
1096         const struct dt_it_ops  *iops   = &parent->do_index_ops->dio_it;
1097         struct dt_it            *it;
1098         int                      rc;
1099         ENTRY;
1100
1101         it = iops->init(env, parent, LUDA_64BITHASH);
1102         if (IS_ERR(it))
1103                 RETURN(PTR_ERR(it));
1104
1105         rc = iops->load(env, it, 0);
1106         if (rc == 0)
1107                 rc = iops->next(env, it);
1108         else if (rc > 0)
1109                 rc = 0;
1110
1111         while (rc == 0) {
1112                 int off = 3;
1113
1114                 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
1115                 if (rc != 0)
1116                         break;
1117
1118                 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
1119                 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
1120                         goto next;
1121
1122                 /* name length must be strlen("MDTxxxx") */
1123                 if (ent->lde_namelen != 7)
1124                         goto remove;
1125
1126                 if (memcmp(ent->lde_name, "MDT", off) != 0)
1127                         goto remove;
1128
1129                 while (off < 7 && isxdigit(ent->lde_name[off]))
1130                         off++;
1131
1132                 if (off != 7) {
1133
1134 remove:
1135                         rc = lfsck_lpf_remove_name_entry(env, lfsck,
1136                                                          ent->lde_name);
1137                         if (rc != 0)
1138                                 break;
1139                 }
1140
1141 next:
1142                 rc = iops->next(env, it);
1143         }
1144
1145         iops->put(env, it);
1146         iops->fini(env, it);
1147
1148         RETURN(rc > 0 ? 0 : rc);
1149 }
1150
1151 static int lfsck_update_lpf_entry(const struct lu_env *env,
1152                                   struct lfsck_instance *lfsck,
1153                                   struct dt_object *parent,
1154                                   struct dt_object *child,
1155                                   const char *name,
1156                                   enum lfsck_verify_lpf_types type)
1157 {
1158         int rc;
1159
1160         if (type == LVLT_BY_BOOKMARK) {
1161                 rc = lfsck_update_name_entry(env, lfsck, parent, name,
1162                                              lfsck_dto2fid(child), S_IFDIR);
1163         } else /* if (type == LVLT_BY_NAMEENTRY) */ {
1164                 lfsck->li_bookmark_ram.lb_lpf_fid = *lfsck_dto2fid(child);
1165                 rc = lfsck_bookmark_store(env, lfsck);
1166
1167                 CDEBUG(D_LFSCK, "%s: update LPF fid "DFID
1168                        " in the bookmark file: rc = %d\n",
1169                        lfsck_lfsck2name(lfsck),
1170                        PFID(lfsck_dto2fid(child)), rc);
1171         }
1172
1173         return rc;
1174 }
1175
1176 /**
1177  * Check whether the @child back references the @parent.
1178  *
1179  * Two cases:
1180  * 1) The child's FID is stored in the bookmark file. If the child back
1181  *    references the parent (LU_LPF_FID object) via its ".." entry, then
1182  *    insert the name (MDTxxxx) to the .lustre/lost+found; otherwise, if
1183  *    the child back references another parent2, then:
1184  * 1.1) If the parent2 recognizes the child, then update the bookmark file;
1185  * 1.2) Otherwise, the LFSCK cannot know whether there will be parent3 that
1186  *      references the child. So keep them there. As the LFSCK processing,
1187  *      the parent3 may be found, then when the LFSCK run next time, the
1188  *      inconsistency can be repaired.
1189  *
1190  * 2) The child's FID is stored in the .lustre/lost+found/ sub-directory name
1191  *    entry (MDTxxxx). If the child back references the parent (LU_LPF_FID obj)
1192  *    via its ".." entry, then update the bookmark file, otherwise, if the child
1193  *    back references another parent2, then:
1194  * 2.1) If the parent2 recognizes the child, then remove the sub-directory
1195  *      from .lustre/lost+found/;
1196  * 2.2) Otherwise, if the parent2 does not recognizes the child, trust the
1197  *      sub-directory name entry and update the child;
1198  * 2.3) Otherwise, if we do not know whether the parent2 recognizes the child
1199  *      or not, then keep them there.
1200  *
1201  * \param[in] env       pointer to the thread context
1202  * \param[in] lfsck     pointer to the lfsck instance
1203  * \param[in] child     pointer to the lost+found sub-directory object
1204  * \param[in] name      the name for lost+found sub-directory object
1205  * \param[out] fid      pointer to the buffer to hold the FID of the object
1206  *                      (called it as parent2) that is referenced via the
1207  *                      child's dotdot entry; it also can be the FID that
1208  *                      is referenced by the name entry under the parent2.
1209  * \param[in] type      to indicate where the child's FID is stored in
1210  *
1211  * \retval              positive number for uncertain inconsistency
1212  * \retval              0 for success
1213  * \retval              negative error number on failure
1214  */
1215 static int lfsck_verify_lpf_pairs(const struct lu_env *env,
1216                                   struct lfsck_instance *lfsck,
1217                                   struct dt_object *child, const char *name,
1218                                   struct lu_fid *fid,
1219                                   enum lfsck_verify_lpf_types type)
1220 {
1221         struct dt_object         *parent  = lfsck->li_lpf_root_obj;
1222         struct lfsck_thread_info *info    = lfsck_env_info(env);
1223         char                     *name2   = info->lti_key;
1224         struct lu_fid            *fid2    = &info->lti_fid3;
1225         struct dt_object         *parent2 = NULL;
1226         struct lustre_handle      lh      = { 0 };
1227         int                       rc;
1228         ENTRY;
1229
1230         fid_zero(fid);
1231         rc = dt_lookup(env, child, (struct dt_rec *)fid,
1232                        (const struct dt_key *)dotdot);
1233         if (rc != 0)
1234                 GOTO(linkea, rc);
1235
1236         if (!fid_is_sane(fid))
1237                 GOTO(linkea, rc = -EINVAL);
1238
1239         if (lu_fid_eq(fid, &LU_LPF_FID)) {
1240                 const struct lu_name *cname;
1241
1242                 if (lfsck->li_lpf_obj == NULL) {
1243                         lu_object_get(&child->do_lu);
1244                         lfsck->li_lpf_obj = child;
1245                 }
1246
1247                 cname = lfsck_name_get_const(env, name, strlen(name));
1248                 rc = lfsck_verify_linkea(env, child, cname, &LU_LPF_FID);
1249                 if (rc == 0)
1250                         rc = lfsck_update_lpf_entry(env, lfsck, parent, child,
1251                                                     name, type);
1252
1253                 GOTO(out_done, rc);
1254         }
1255
1256         parent2 = lfsck_object_find_bottom(env, lfsck, fid);
1257         if (IS_ERR(parent2))
1258                 GOTO(linkea, parent2);
1259
1260         if (!dt_object_exists(parent2)) {
1261                 lfsck_object_put(env, parent2);
1262
1263                 GOTO(linkea, parent2 = ERR_PTR(-ENOENT));
1264         }
1265
1266         if (!dt_try_as_dir(env, parent2)) {
1267                 lfsck_object_put(env, parent2);
1268
1269                 GOTO(linkea, parent2 = ERR_PTR(-ENOTDIR));
1270         }
1271
1272 linkea:
1273         /* To prevent rename/unlink race */
1274         rc = lfsck_ibits_lock(env, lfsck, child, &lh,
1275                               MDS_INODELOCK_UPDATE, LCK_PR);
1276         if (rc != 0)
1277                 GOTO(out_put, rc);
1278
1279         dt_read_lock(env, child, 0);
1280         rc = lfsck_links_get_first(env, child, name2, fid2);
1281         if (rc != 0) {
1282                 dt_read_unlock(env, child);
1283                 lfsck_ibits_unlock(&lh, LCK_PR);
1284
1285                 GOTO(out_put, rc = 1);
1286         }
1287
1288         /* It is almost impossible that the bookmark file (or the name entry)
1289          * and the linkEA hit the same data corruption. Trust the linkEA. */
1290         if (lu_fid_eq(fid2, &LU_LPF_FID) && strcmp(name, name2) == 0) {
1291                 dt_read_unlock(env, child);
1292                 lfsck_ibits_unlock(&lh, LCK_PR);
1293
1294                 *fid = *fid2;
1295                 if (lfsck->li_lpf_obj == NULL) {
1296                         lu_object_get(&child->do_lu);
1297                         lfsck->li_lpf_obj = child;
1298                 }
1299
1300                 /* Update the child's dotdot entry */
1301                 rc = lfsck_update_name_entry(env, lfsck, child, dotdot,
1302                                              &LU_LPF_FID, S_IFDIR);
1303                 if (rc == 0)
1304                         rc = lfsck_update_lpf_entry(env, lfsck, parent, child,
1305                                                     name, type);
1306
1307                 GOTO(out_put, rc);
1308         }
1309
1310         if (parent2 == NULL || IS_ERR(parent2)) {
1311                 dt_read_unlock(env, child);
1312                 lfsck_ibits_unlock(&lh, LCK_PR);
1313
1314                 GOTO(out_done, rc = 1);
1315         }
1316
1317         rc = dt_lookup(env, parent2, (struct dt_rec *)fid,
1318                        (const struct dt_key *)name2);
1319         dt_read_unlock(env, child);
1320         lfsck_ibits_unlock(&lh, LCK_PR);
1321         if (rc != 0 && rc != -ENOENT)
1322                 GOTO(out_put, rc);
1323
1324         if (rc == -ENOENT || !lu_fid_eq(fid, lfsck_dto2fid(child))) {
1325                 if (type == LVLT_BY_BOOKMARK)
1326                         GOTO(out_put, rc = 1);
1327
1328                 /* Trust the name entry, update the child's dotdot entry. */
1329                 rc = lfsck_update_name_entry(env, lfsck, child, dotdot,
1330                                              &LU_LPF_FID, S_IFDIR);
1331
1332                 GOTO(out_put, rc);
1333         }
1334
1335         if (type == LVLT_BY_BOOKMARK) {
1336                 /* Invalid FID record in the bookmark file, reset it. */
1337                 fid_zero(&lfsck->li_bookmark_ram.lb_lpf_fid);
1338                 rc = lfsck_bookmark_store(env, lfsck);
1339
1340                 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1341                        " in the bookmark file: rc = %d\n",
1342                        lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(child)), rc);
1343         } else /* if (type == LVLT_BY_NAMEENTRY) */ {
1344                 /* The name entry is wrong, remove it. */
1345                 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1346         }
1347
1348         GOTO(out_put, rc);
1349
1350 out_put:
1351         if (parent2 != NULL && !IS_ERR(parent2))
1352                 lfsck_object_put(env, parent2);
1353
1354 out_done:
1355         return rc;
1356 }
1357
1358 /**
1359  * Verify the /ROOT/.lustre/lost+found/ directory.
1360  *
1361  * /ROOT/.lustre/lost+found/ is a special directory to hold the objects that
1362  * the LFSCK does not exactly know how to handle, such as orphans. So before
1363  * the LFSCK scanning the system, the consistency of such directory needs to
1364  * be verified firstly to allow the users to use it during the LFSCK.
1365  *
1366  * \param[in] env       pointer to the thread context
1367  * \param[in] lfsck     pointer to the lfsck instance
1368  *
1369  * \retval              positive number for uncertain inconsistency
1370  * \retval              0 for success
1371  * \retval              negative error number on failure
1372  */
1373 int lfsck_verify_lpf(const struct lu_env *env, struct lfsck_instance *lfsck)
1374 {
1375         struct lfsck_thread_info *info   = lfsck_env_info(env);
1376         struct lu_fid            *pfid   = &info->lti_fid;
1377         struct lu_fid            *cfid   = &info->lti_fid2;
1378         struct lfsck_bookmark    *bk     = &lfsck->li_bookmark_ram;
1379         struct dt_object         *parent;
1380         /* child1's FID is in the bookmark file. */
1381         struct dt_object         *child1 = NULL;
1382         /* child2's FID is in the name entry MDTxxxx. */
1383         struct dt_object         *child2 = NULL;
1384         const struct lu_name     *cname;
1385         char                      name[8];
1386         int                       node   = lfsck_dev_idx(lfsck);
1387         int                       rc     = 0;
1388         ENTRY;
1389
1390         LASSERT(lfsck->li_master);
1391
1392         if (lfsck->li_lpf_root_obj != NULL)
1393                 RETURN(0);
1394
1395         if (node == 0) {
1396                 parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
1397                                                   &LU_LPF_FID);
1398         } else {
1399                 struct lfsck_tgt_desc *ltd;
1400
1401                 ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, 0);
1402                 if (unlikely(ltd == NULL))
1403                         RETURN(-ENXIO);
1404
1405                 parent = lfsck_object_find_by_dev(env, ltd->ltd_tgt,
1406                                                   &LU_LPF_FID);
1407                 lfsck_tgt_put(ltd);
1408         }
1409
1410         if (IS_ERR(parent))
1411                 RETURN(PTR_ERR(parent));
1412
1413         LASSERT(dt_object_exists(parent));
1414
1415         if (unlikely(!dt_try_as_dir(env, parent))) {
1416                 lfsck_object_put(env, parent);
1417
1418                 GOTO(put, rc = -ENOTDIR);
1419         }
1420
1421         lfsck->li_lpf_root_obj = parent;
1422         if (node == 0) {
1423                 rc = lfsck_scan_lpf_bad_entries(env, lfsck);
1424                 if (rc != 0)
1425                         CDEBUG(D_LFSCK, "%s: scan .lustre/lost+found/ "
1426                                "for bad sub-directories: rc = %d\n",
1427                                lfsck_lfsck2name(lfsck), rc);
1428         }
1429
1430         /* child2 */
1431         snprintf(name, 8, "MDT%04x", node);
1432         rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1433                        (const struct dt_key *)name);
1434         if (rc == -ENOENT) {
1435                 rc = 0;
1436                 goto find_child1;
1437         }
1438
1439         if (rc != 0)
1440                 GOTO(put, rc);
1441
1442         /* Invalid FID in the name entry, remove the name entry. */
1443         if (!fid_is_norm(cfid)) {
1444                 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1445                 if (rc != 0)
1446                         GOTO(put, rc);
1447
1448                 goto find_child1;
1449         }
1450
1451         child2 = lfsck_object_find_bottom(env, lfsck, cfid);
1452         if (IS_ERR(child2))
1453                 GOTO(put, rc = PTR_ERR(child2));
1454
1455         if (unlikely(!dt_object_exists(child2) ||
1456                      dt_object_remote(child2)) ||
1457                      !S_ISDIR(lfsck_object_type(child2))) {
1458                 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1459                 if (rc != 0)
1460                         GOTO(put, rc);
1461
1462                 goto find_child1;
1463         }
1464
1465         if (unlikely(!dt_try_as_dir(env, child2))) {
1466                 lfsck_object_put(env, child2);
1467                 child2 = NULL;
1468                 rc = -ENOTDIR;
1469         }
1470
1471 find_child1:
1472         if (fid_is_zero(&bk->lb_lpf_fid))
1473                 goto check_child2;
1474
1475         if (likely(lu_fid_eq(cfid, &bk->lb_lpf_fid))) {
1476                 if (lfsck->li_lpf_obj == NULL) {
1477                         lu_object_get(&child2->do_lu);
1478                         lfsck->li_lpf_obj = child2;
1479                 }
1480
1481                 cname = lfsck_name_get_const(env, name, strlen(name));
1482                 rc = lfsck_verify_linkea(env, child2, cname, &LU_LPF_FID);
1483
1484                 GOTO(put, rc);
1485         }
1486
1487         if (unlikely(!fid_is_norm(&bk->lb_lpf_fid))) {
1488                 struct lu_fid tfid = bk->lb_lpf_fid;
1489
1490                 /* Invalid FID record in the bookmark file, reset it. */
1491                 fid_zero(&bk->lb_lpf_fid);
1492                 rc = lfsck_bookmark_store(env, lfsck);
1493
1494                 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1495                        " in the bookmark file: rc = %d\n",
1496                        lfsck_lfsck2name(lfsck), PFID(&tfid), rc);
1497
1498                 if (rc != 0)
1499                         GOTO(put, rc);
1500
1501                 goto check_child2;
1502         }
1503
1504         child1 = lfsck_object_find_bottom(env, lfsck, &bk->lb_lpf_fid);
1505         if (IS_ERR(child1)) {
1506                 child1 = NULL;
1507                 goto check_child2;
1508         }
1509
1510         if (unlikely(!dt_object_exists(child1) ||
1511                      dt_object_remote(child1)) ||
1512                      !S_ISDIR(lfsck_object_type(child1))) {
1513                 /* Invalid FID record in the bookmark file, reset it. */
1514                 fid_zero(&bk->lb_lpf_fid);
1515                 rc = lfsck_bookmark_store(env, lfsck);
1516
1517                 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1518                        " in the bookmark file: rc = %d\n",
1519                        lfsck_lfsck2name(lfsck),
1520                        PFID(lfsck_dto2fid(child1)), rc);
1521
1522                 if (rc != 0)
1523                         GOTO(put, rc);
1524
1525                 lfsck_object_put(env, child1);
1526                 child1 = NULL;
1527                 goto check_child2;
1528         }
1529
1530         if (unlikely(!dt_try_as_dir(env, child1))) {
1531                 lfsck_object_put(env, child1);
1532                 child1 = NULL;
1533                 rc = -ENOTDIR;
1534                 goto check_child2;
1535         }
1536
1537         rc = lfsck_verify_lpf_pairs(env, lfsck, child1, name, pfid,
1538                                     LVLT_BY_BOOKMARK);
1539         if (lu_fid_eq(pfid, &LU_LPF_FID))
1540                 GOTO(put, rc);
1541
1542 check_child2:
1543         if (child2 != NULL)
1544                 rc = lfsck_verify_lpf_pairs(env, lfsck, child2, name,
1545                                             pfid, LVLT_BY_NAMEENTRY);
1546
1547         GOTO(put, rc);
1548
1549 put:
1550         if (lfsck->li_lpf_obj != NULL) {
1551                 if (unlikely(!dt_try_as_dir(env, lfsck->li_lpf_obj))) {
1552                         lfsck_object_put(env, lfsck->li_lpf_obj);
1553                         lfsck->li_lpf_obj = NULL;
1554                         rc = -ENOTDIR;
1555                 }
1556         } else if (rc == 0) {
1557                 rc = lfsck_create_lpf(env, lfsck);
1558         }
1559
1560         if (child2 != NULL && !IS_ERR(child2))
1561                 lfsck_object_put(env, child2);
1562         if (child1 != NULL && !IS_ERR(child1))
1563                 lfsck_object_put(env, child1);
1564
1565         return rc;
1566 }
1567
1568 static int lfsck_fid_init(struct lfsck_instance *lfsck)
1569 {
1570         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
1571         struct seq_server_site  *ss     = lfsck_dev_site(lfsck);
1572         char                    *prefix;
1573         int                      rc     = 0;
1574         ENTRY;
1575
1576         if (unlikely(ss == NULL))
1577                 RETURN(-ENXIO);
1578
1579         OBD_ALLOC_PTR(lfsck->li_seq);
1580         if (lfsck->li_seq == NULL)
1581                 RETURN(-ENOMEM);
1582
1583         OBD_ALLOC(prefix, MAX_OBD_NAME + 7);
1584         if (prefix == NULL)
1585                 GOTO(out, rc = -ENOMEM);
1586
1587         snprintf(prefix, MAX_OBD_NAME + 7, "lfsck-%s", lfsck_lfsck2name(lfsck));
1588         rc = seq_client_init(lfsck->li_seq, NULL, LUSTRE_SEQ_METADATA, prefix,
1589                              ss->ss_server_seq);
1590         OBD_FREE(prefix, MAX_OBD_NAME + 7);
1591         if (rc != 0)
1592                 GOTO(out, rc);
1593
1594         if (fid_is_sane(&bk->lb_last_fid))
1595                 lfsck->li_seq->lcs_fid = bk->lb_last_fid;
1596
1597         RETURN(0);
1598
1599 out:
1600         OBD_FREE_PTR(lfsck->li_seq);
1601         lfsck->li_seq = NULL;
1602
1603         return rc;
1604 }
1605
1606 static void lfsck_fid_fini(struct lfsck_instance *lfsck)
1607 {
1608         if (lfsck->li_seq != NULL) {
1609                 seq_client_fini(lfsck->li_seq);
1610                 OBD_FREE_PTR(lfsck->li_seq);
1611                 lfsck->li_seq = NULL;
1612         }
1613 }
1614
1615 void lfsck_instance_cleanup(const struct lu_env *env,
1616                             struct lfsck_instance *lfsck)
1617 {
1618         struct ptlrpc_thread    *thread = &lfsck->li_thread;
1619         struct lfsck_component  *com;
1620         struct lfsck_component  *next;
1621         struct lfsck_lmv_unit   *llu;
1622         struct lfsck_lmv_unit   *llu_next;
1623         struct lfsck_lmv        *llmv;
1624         ENTRY;
1625
1626         LASSERT(list_empty(&lfsck->li_link));
1627         LASSERT(thread_is_init(thread) || thread_is_stopped(thread));
1628
1629         if (lfsck->li_obj_oit != NULL) {
1630                 lfsck_object_put(env, lfsck->li_obj_oit);
1631                 lfsck->li_obj_oit = NULL;
1632         }
1633
1634         LASSERT(lfsck->li_obj_dir == NULL);
1635         LASSERT(lfsck->li_lmv == NULL);
1636
1637         list_for_each_entry_safe(llu, llu_next, &lfsck->li_list_lmv, llu_link) {
1638                 llmv = &llu->llu_lmv;
1639
1640                 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
1641                          "still in using: %u\n",
1642                          atomic_read(&llmv->ll_ref));
1643
1644                 lfsck_lmv_put(env, llmv);
1645         }
1646
1647         list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
1648                 lfsck_component_cleanup(env, com);
1649         }
1650
1651         LASSERT(list_empty(&lfsck->li_list_dir));
1652
1653         list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
1654                                  lc_link) {
1655                 lfsck_component_cleanup(env, com);
1656         }
1657
1658         list_for_each_entry_safe(com, next, &lfsck->li_list_idle, lc_link) {
1659                 lfsck_component_cleanup(env, com);
1660         }
1661
1662         lfsck_tgt_descs_fini(&lfsck->li_ost_descs);
1663         lfsck_tgt_descs_fini(&lfsck->li_mdt_descs);
1664
1665         if (lfsck->li_lfsck_dir != NULL) {
1666                 lfsck_object_put(env, lfsck->li_lfsck_dir);
1667                 lfsck->li_lfsck_dir = NULL;
1668         }
1669
1670         if (lfsck->li_bookmark_obj != NULL) {
1671                 lfsck_object_put(env, lfsck->li_bookmark_obj);
1672                 lfsck->li_bookmark_obj = NULL;
1673         }
1674
1675         if (lfsck->li_lpf_obj != NULL) {
1676                 lfsck_object_put(env, lfsck->li_lpf_obj);
1677                 lfsck->li_lpf_obj = NULL;
1678         }
1679
1680         if (lfsck->li_lpf_root_obj != NULL) {
1681                 lfsck_object_put(env, lfsck->li_lpf_root_obj);
1682                 lfsck->li_lpf_root_obj = NULL;
1683         }
1684
1685         if (lfsck->li_los != NULL) {
1686                 local_oid_storage_fini(env, lfsck->li_los);
1687                 lfsck->li_los = NULL;
1688         }
1689
1690         lfsck_fid_fini(lfsck);
1691
1692         OBD_FREE_PTR(lfsck);
1693 }
1694
1695 static inline struct lfsck_instance *
1696 __lfsck_instance_find(struct dt_device *key, bool ref, bool unlink)
1697 {
1698         struct lfsck_instance *lfsck;
1699
1700         list_for_each_entry(lfsck, &lfsck_instance_list, li_link) {
1701                 if (lfsck->li_bottom == key) {
1702                         if (ref)
1703                                 lfsck_instance_get(lfsck);
1704                         if (unlink)
1705                                 list_del_init(&lfsck->li_link);
1706
1707                         return lfsck;
1708                 }
1709         }
1710
1711         return NULL;
1712 }
1713
1714 struct lfsck_instance *lfsck_instance_find(struct dt_device *key, bool ref,
1715                                            bool unlink)
1716 {
1717         struct lfsck_instance *lfsck;
1718
1719         spin_lock(&lfsck_instance_lock);
1720         lfsck = __lfsck_instance_find(key, ref, unlink);
1721         spin_unlock(&lfsck_instance_lock);
1722
1723         return lfsck;
1724 }
1725
1726 static inline int lfsck_instance_add(struct lfsck_instance *lfsck)
1727 {
1728         struct lfsck_instance *tmp;
1729
1730         spin_lock(&lfsck_instance_lock);
1731         list_for_each_entry(tmp, &lfsck_instance_list, li_link) {
1732                 if (lfsck->li_bottom == tmp->li_bottom) {
1733                         spin_unlock(&lfsck_instance_lock);
1734                         return -EEXIST;
1735                 }
1736         }
1737
1738         list_add_tail(&lfsck->li_link, &lfsck_instance_list);
1739         spin_unlock(&lfsck_instance_lock);
1740         return 0;
1741 }
1742
1743 void lfsck_bits_dump(struct seq_file *m, int bits, const char *names[],
1744                      const char *prefix)
1745 {
1746         int flag;
1747         int i;
1748         bool newline = (bits != 0 ? false : true);
1749
1750         seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
1751
1752         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
1753                 if (flag & bits) {
1754                         bits &= ~flag;
1755                         if (names[i] != NULL) {
1756                                 if (bits == 0)
1757                                         newline = true;
1758
1759                                 seq_printf(m, "%s%c", names[i],
1760                                            newline ? '\n' : ',');
1761                         }
1762                 }
1763         }
1764
1765         if (!newline)
1766                 seq_putc(m, '\n');
1767 }
1768
1769 void lfsck_time_dump(struct seq_file *m, time64_t time, const char *name)
1770 {
1771         if (time == 0) {
1772                 seq_printf(m, "%s_time: N/A\n", name);
1773                 seq_printf(m, "time_since_%s: N/A\n", name);
1774         } else {
1775                 seq_printf(m, "%s_time: %lld\n", name, time);
1776                 seq_printf(m, "time_since_%s: %lld seconds\n",
1777                            name, ktime_get_real_seconds() - time);
1778         }
1779 }
1780
1781 void lfsck_pos_dump(struct seq_file *m, struct lfsck_position *pos,
1782                     const char *prefix)
1783 {
1784         if (fid_is_zero(&pos->lp_dir_parent)) {
1785                 if (pos->lp_oit_cookie == 0) {
1786                         seq_printf(m, "%s: N/A, N/A, N/A\n", prefix);
1787                         return;
1788                 }
1789                 seq_printf(m, "%s: %llu, N/A, N/A\n",
1790                            prefix, pos->lp_oit_cookie);
1791         } else {
1792                 seq_printf(m, "%s: %llu, "DFID", %#llx\n",
1793                            prefix, pos->lp_oit_cookie,
1794                            PFID(&pos->lp_dir_parent), pos->lp_dir_cookie);
1795         }
1796 }
1797
1798 void lfsck_pos_fill(const struct lu_env *env, struct lfsck_instance *lfsck,
1799                     struct lfsck_position *pos, bool init)
1800 {
1801         const struct dt_it_ops *iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
1802
1803         if (unlikely(lfsck->li_di_oit == NULL)) {
1804                 memset(pos, 0, sizeof(*pos));
1805                 return;
1806         }
1807
1808         pos->lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
1809         if (!lfsck->li_current_oit_processed && !init)
1810                 pos->lp_oit_cookie--;
1811
1812         LASSERT(pos->lp_oit_cookie > 0);
1813
1814         if (lfsck->li_di_dir != NULL) {
1815                 struct dt_object *dto = lfsck->li_obj_dir;
1816
1817                 pos->lp_dir_cookie = dto->do_index_ops->dio_it.store(env,
1818                                                         lfsck->li_di_dir);
1819
1820                 if (pos->lp_dir_cookie >= MDS_DIR_END_OFF) {
1821                         fid_zero(&pos->lp_dir_parent);
1822                         pos->lp_dir_cookie = 0;
1823                 } else {
1824                         pos->lp_dir_parent = *lfsck_dto2fid(dto);
1825                 }
1826         } else {
1827                 fid_zero(&pos->lp_dir_parent);
1828                 pos->lp_dir_cookie = 0;
1829         }
1830 }
1831
1832 bool __lfsck_set_speed(struct lfsck_instance *lfsck, __u32 limit)
1833 {
1834         bool dirty = false;
1835
1836         if (limit != LFSCK_SPEED_NO_LIMIT) {
1837                 if (limit > msecs_to_jiffies(MSEC_PER_SEC)) {
1838                         lfsck->li_sleep_rate = limit /
1839                                                msecs_to_jiffies(MSEC_PER_SEC);
1840                         lfsck->li_sleep_jif = 1;
1841                 } else {
1842                         lfsck->li_sleep_rate = 1;
1843                         lfsck->li_sleep_jif = msecs_to_jiffies(MSEC_PER_SEC) /
1844                                               limit;
1845                 }
1846         } else {
1847                 lfsck->li_sleep_jif = 0;
1848                 lfsck->li_sleep_rate = 0;
1849         }
1850
1851         if (lfsck->li_bookmark_ram.lb_speed_limit != limit) {
1852                 lfsck->li_bookmark_ram.lb_speed_limit = limit;
1853                 dirty = true;
1854         }
1855
1856         return dirty;
1857 }
1858
1859 void lfsck_control_speed(struct lfsck_instance *lfsck)
1860 {
1861         struct ptlrpc_thread *thread = &lfsck->li_thread;
1862         struct l_wait_info    lwi;
1863
1864         if (lfsck->li_sleep_jif > 0 &&
1865             lfsck->li_new_scanned >= lfsck->li_sleep_rate) {
1866                 lwi = LWI_TIMEOUT_INTR(lfsck->li_sleep_jif, NULL,
1867                                        LWI_ON_SIGNAL_NOOP, NULL);
1868
1869                 l_wait_event(thread->t_ctl_waitq,
1870                              !thread_is_running(thread),
1871                              &lwi);
1872                 lfsck->li_new_scanned = 0;
1873         }
1874 }
1875
1876 void lfsck_control_speed_by_self(struct lfsck_component *com)
1877 {
1878         struct lfsck_instance   *lfsck  = com->lc_lfsck;
1879         struct ptlrpc_thread    *thread = &lfsck->li_thread;
1880         struct l_wait_info       lwi;
1881
1882         if (lfsck->li_sleep_jif > 0 &&
1883             com->lc_new_scanned >= lfsck->li_sleep_rate) {
1884                 lwi = LWI_TIMEOUT_INTR(lfsck->li_sleep_jif, NULL,
1885                                        LWI_ON_SIGNAL_NOOP, NULL);
1886
1887                 l_wait_event(thread->t_ctl_waitq,
1888                              !thread_is_running(thread),
1889                              &lwi);
1890                 com->lc_new_scanned = 0;
1891         }
1892 }
1893
1894 static struct lfsck_thread_args *
1895 lfsck_thread_args_init(struct lfsck_instance *lfsck,
1896                        struct lfsck_component *com,
1897                        struct lfsck_start_param *lsp)
1898 {
1899         struct lfsck_thread_args *lta;
1900         int                       rc;
1901
1902         OBD_ALLOC_PTR(lta);
1903         if (lta == NULL)
1904                 return ERR_PTR(-ENOMEM);
1905
1906         rc = lu_env_init(&lta->lta_env, LCT_MD_THREAD | LCT_DT_THREAD);
1907         if (rc != 0) {
1908                 OBD_FREE_PTR(lta);
1909                 return ERR_PTR(rc);
1910         }
1911
1912         lta->lta_lfsck = lfsck_instance_get(lfsck);
1913         if (com != NULL)
1914                 lta->lta_com = lfsck_component_get(com);
1915
1916         lta->lta_lsp = lsp;
1917
1918         return lta;
1919 }
1920
1921 void lfsck_thread_args_fini(struct lfsck_thread_args *lta)
1922 {
1923         if (lta->lta_com != NULL)
1924                 lfsck_component_put(&lta->lta_env, lta->lta_com);
1925         lfsck_instance_put(&lta->lta_env, lta->lta_lfsck);
1926         lu_env_fini(&lta->lta_env);
1927         OBD_FREE_PTR(lta);
1928 }
1929
1930 struct lfsck_assistant_data *
1931 lfsck_assistant_data_init(struct lfsck_assistant_operations *lao,
1932                           const char *name)
1933 {
1934         struct lfsck_assistant_data *lad;
1935
1936         OBD_ALLOC_PTR(lad);
1937         if (lad != NULL) {
1938                 lad->lad_bitmap = CFS_ALLOCATE_BITMAP(BITS_PER_LONG);
1939                 if (lad->lad_bitmap == NULL) {
1940                         OBD_FREE_PTR(lad);
1941                         return NULL;
1942                 }
1943
1944                 INIT_LIST_HEAD(&lad->lad_req_list);
1945                 spin_lock_init(&lad->lad_lock);
1946                 INIT_LIST_HEAD(&lad->lad_ost_list);
1947                 INIT_LIST_HEAD(&lad->lad_ost_phase1_list);
1948                 INIT_LIST_HEAD(&lad->lad_ost_phase2_list);
1949                 INIT_LIST_HEAD(&lad->lad_mdt_list);
1950                 INIT_LIST_HEAD(&lad->lad_mdt_phase1_list);
1951                 INIT_LIST_HEAD(&lad->lad_mdt_phase2_list);
1952                 init_waitqueue_head(&lad->lad_thread.t_ctl_waitq);
1953                 lad->lad_ops = lao;
1954                 lad->lad_name = name;
1955         }
1956
1957         return lad;
1958 }
1959
1960 struct lfsck_assistant_object *
1961 lfsck_assistant_object_init(const struct lu_env *env, const struct lu_fid *fid,
1962                             const struct lu_attr *attr, __u64 cookie,
1963                             bool is_dir)
1964 {
1965         struct lfsck_assistant_object   *lso;
1966
1967         OBD_ALLOC_PTR(lso);
1968         if (lso == NULL)
1969                 return ERR_PTR(-ENOMEM);
1970
1971         lso->lso_fid = *fid;
1972         if (attr != NULL)
1973                 lso->lso_attr = *attr;
1974
1975         atomic_set(&lso->lso_ref, 1);
1976         lso->lso_oit_cookie = cookie;
1977         if (is_dir)
1978                 lso->lso_is_dir = 1;
1979
1980         return lso;
1981 }
1982
1983 struct dt_object *
1984 lfsck_assistant_object_load(const struct lu_env *env,
1985                             struct lfsck_instance *lfsck,
1986                             struct lfsck_assistant_object *lso)
1987 {
1988         struct dt_object *obj;
1989
1990         obj = lfsck_object_find_bottom(env, lfsck, &lso->lso_fid);
1991         if (IS_ERR(obj))
1992                 return obj;
1993
1994         if (unlikely(!dt_object_exists(obj) || lfsck_is_dead_obj(obj))) {
1995                 lso->lso_dead = 1;
1996                 lfsck_object_put(env, obj);
1997
1998                 return ERR_PTR(-ENOENT);
1999         }
2000
2001         if (lso->lso_is_dir && unlikely(!dt_try_as_dir(env, obj))) {
2002                 lfsck_object_put(env, obj);
2003
2004                 return ERR_PTR(-ENOTDIR);
2005         }
2006
2007         return obj;
2008 }
2009
2010 /**
2011  * Generic LFSCK asynchronous communication interpretor function.
2012  * The LFSCK RPC reply for both the event notification and status
2013  * querying will be handled here.
2014  *
2015  * \param[in] env       pointer to the thread context
2016  * \param[in] req       pointer to the LFSCK request
2017  * \param[in] args      pointer to the lfsck_async_interpret_args
2018  * \param[in] rc        the result for handling the LFSCK request
2019  *
2020  * \retval              0 for success
2021  * \retval              negative error number on failure
2022  */
2023 int lfsck_async_interpret_common(const struct lu_env *env,
2024                                  struct ptlrpc_request *req,
2025                                  void *args, int rc)
2026 {
2027         struct lfsck_async_interpret_args *laia = args;
2028         struct lfsck_component            *com  = laia->laia_com;
2029         struct lfsck_assistant_data       *lad  = com->lc_data;
2030         struct lfsck_tgt_descs            *ltds = laia->laia_ltds;
2031         struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
2032         struct lfsck_request              *lr   = laia->laia_lr;
2033
2034         LASSERT(com->lc_lfsck->li_master);
2035
2036         switch (lr->lr_event) {
2037         case LE_START:
2038                 if (unlikely(rc == -EINPROGRESS)) {
2039                         ltd->ltd_retry_start = 1;
2040                         break;
2041                 }
2042
2043                 if (rc != 0) {
2044                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for %s "
2045                                "start: rc = %d\n",
2046                                lfsck_lfsck2name(com->lc_lfsck),
2047                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2048                                ltd->ltd_index, lad->lad_name, rc);
2049
2050                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2051                                 struct lfsck_layout *lo = com->lc_file_ram;
2052
2053                                 if (lr->lr_flags & LEF_TO_OST)
2054                                         lfsck_lad_set_bitmap(env, com,
2055                                                              ltd->ltd_index);
2056                                 else
2057                                         lo->ll_flags |= LF_INCOMPLETE;
2058                         } else {
2059                                 struct lfsck_namespace *ns = com->lc_file_ram;
2060
2061                                 /* If some MDT does not join the namespace
2062                                  * LFSCK, then we cannot know whether there
2063                                  * is some name entry on such MDT that with
2064                                  * the referenced MDT-object on this MDT or
2065                                  * not. So the namespace LFSCK on this MDT
2066                                  * cannot handle orphan MDT-objects properly.
2067                                  * So we mark the LFSCK as LF_INCOMPLETE and
2068                                  * skip orphan MDT-objects handling. */
2069                                 ns->ln_flags |= LF_INCOMPLETE;
2070                         }
2071                         break;
2072                 }
2073
2074                 spin_lock(&ltds->ltd_lock);
2075                 if (ltd->ltd_dead) {
2076                         spin_unlock(&ltds->ltd_lock);
2077                         break;
2078                 }
2079
2080                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2081                         struct list_head *list;
2082                         struct list_head *phase_list;
2083
2084                         if (ltd->ltd_layout_done) {
2085                                 spin_unlock(&ltds->ltd_lock);
2086                                 break;
2087                         }
2088
2089                         if (lr->lr_flags & LEF_TO_OST) {
2090                                 list = &lad->lad_ost_list;
2091                                 phase_list = &lad->lad_ost_phase1_list;
2092                         } else {
2093                                 list = &lad->lad_mdt_list;
2094                                 phase_list = &lad->lad_mdt_phase1_list;
2095                         }
2096
2097                         if (list_empty(&ltd->ltd_layout_list))
2098                                 list_add_tail(&ltd->ltd_layout_list, list);
2099                         if (list_empty(&ltd->ltd_layout_phase_list))
2100                                 list_add_tail(&ltd->ltd_layout_phase_list,
2101                                               phase_list);
2102                 } else {
2103                         if (ltd->ltd_namespace_done) {
2104                                 spin_unlock(&ltds->ltd_lock);
2105                                 break;
2106                         }
2107
2108                         if (list_empty(&ltd->ltd_namespace_list))
2109                                 list_add_tail(&ltd->ltd_namespace_list,
2110                                               &lad->lad_mdt_list);
2111                         if (list_empty(&ltd->ltd_namespace_phase_list))
2112                                 list_add_tail(&ltd->ltd_namespace_phase_list,
2113                                               &lad->lad_mdt_phase1_list);
2114                 }
2115                 spin_unlock(&ltds->ltd_lock);
2116                 break;
2117         case LE_STOP:
2118         case LE_PHASE1_DONE:
2119         case LE_PHASE2_DONE:
2120         case LE_PEER_EXIT:
2121                 if (rc != 0 && rc != -EALREADY)
2122                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for %s: "
2123                               "event = %d, rc = %d\n",
2124                               lfsck_lfsck2name(com->lc_lfsck),
2125                               (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2126                               ltd->ltd_index, lad->lad_name, lr->lr_event, rc);
2127                 break;
2128         case LE_QUERY: {
2129                 struct lfsck_reply *reply;
2130                 struct list_head *list;
2131                 struct list_head *phase_list;
2132
2133                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2134                         list = &ltd->ltd_layout_list;
2135                         phase_list = &ltd->ltd_layout_phase_list;
2136                 } else {
2137                         list = &ltd->ltd_namespace_list;
2138                         phase_list = &ltd->ltd_namespace_phase_list;
2139                 }
2140
2141                 if (rc != 0) {
2142                         if (lr->lr_flags & LEF_QUERY_ALL) {
2143                                 lfsck_reset_ltd_status(ltd, com->lc_type);
2144                                 break;
2145                         }
2146
2147                         spin_lock(&ltds->ltd_lock);
2148                         list_del_init(phase_list);
2149                         list_del_init(list);
2150                         spin_unlock(&ltds->ltd_lock);
2151                         break;
2152                 }
2153
2154                 reply = req_capsule_server_get(&req->rq_pill,
2155                                                &RMF_LFSCK_REPLY);
2156                 if (reply == NULL) {
2157                         rc = -EPROTO;
2158                         CDEBUG(D_LFSCK, "%s: invalid query reply for %s: "
2159                                "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
2160                                lad->lad_name, rc);
2161
2162                         if (lr->lr_flags & LEF_QUERY_ALL) {
2163                                 lfsck_reset_ltd_status(ltd, com->lc_type);
2164                                 break;
2165                         }
2166
2167                         spin_lock(&ltds->ltd_lock);
2168                         list_del_init(phase_list);
2169                         list_del_init(list);
2170                         spin_unlock(&ltds->ltd_lock);
2171                         break;
2172                 }
2173
2174                 if (lr->lr_flags & LEF_QUERY_ALL) {
2175                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2176                                 ltd->ltd_layout_status = reply->lr_status;
2177                                 ltd->ltd_layout_repaired = reply->lr_repaired;
2178                         } else {
2179                                 ltd->ltd_namespace_status = reply->lr_status;
2180                                 ltd->ltd_namespace_repaired =
2181                                                         reply->lr_repaired;
2182                         }
2183                         break;
2184                 }
2185
2186                 switch (reply->lr_status) {
2187                 case LS_SCANNING_PHASE1:
2188                         break;
2189                 case LS_SCANNING_PHASE2:
2190                         spin_lock(&ltds->ltd_lock);
2191                         list_del_init(phase_list);
2192                         if (ltd->ltd_dead) {
2193                                 spin_unlock(&ltds->ltd_lock);
2194                                 break;
2195                         }
2196
2197                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2198                                 if (ltd->ltd_layout_done) {
2199                                         spin_unlock(&ltds->ltd_lock);
2200                                         break;
2201                                 }
2202
2203                                 if (lr->lr_flags & LEF_TO_OST)
2204                                         list_add_tail(phase_list,
2205                                                 &lad->lad_ost_phase2_list);
2206                                 else
2207                                         list_add_tail(phase_list,
2208                                                 &lad->lad_mdt_phase2_list);
2209                         } else {
2210                                 if (ltd->ltd_namespace_done) {
2211                                         spin_unlock(&ltds->ltd_lock);
2212                                         break;
2213                                 }
2214
2215                                 list_add_tail(phase_list,
2216                                               &lad->lad_mdt_phase2_list);
2217                         }
2218                         spin_unlock(&ltds->ltd_lock);
2219                         break;
2220                 default:
2221                         spin_lock(&ltds->ltd_lock);
2222                         list_del_init(phase_list);
2223                         list_del_init(list);
2224                         spin_unlock(&ltds->ltd_lock);
2225                         break;
2226                 }
2227                 break;
2228         }
2229         default:
2230                 CDEBUG(D_LFSCK, "%s: unexpected event: rc = %d\n",
2231                        lfsck_lfsck2name(com->lc_lfsck), lr->lr_event);
2232                 break;
2233         }
2234
2235         if (!laia->laia_shared) {
2236                 lfsck_tgt_put(ltd);
2237                 lfsck_component_put(env, com);
2238         }
2239
2240         return 0;
2241 }
2242
2243 static void lfsck_interpret(const struct lu_env *env,
2244                             struct lfsck_instance *lfsck,
2245                             struct ptlrpc_request *req, void *args, int result)
2246 {
2247         struct lfsck_async_interpret_args *laia = args;
2248         struct lfsck_component            *com;
2249
2250         LASSERT(laia->laia_com == NULL);
2251         LASSERT(laia->laia_shared);
2252
2253         spin_lock(&lfsck->li_lock);
2254         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
2255                 laia->laia_com = com;
2256                 lfsck_async_interpret_common(env, req, laia, result);
2257         }
2258
2259         list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
2260                 laia->laia_com = com;
2261                 lfsck_async_interpret_common(env, req, laia, result);
2262         }
2263         spin_unlock(&lfsck->li_lock);
2264 }
2265
2266 static int lfsck_stop_notify(const struct lu_env *env,
2267                              struct lfsck_instance *lfsck,
2268                              struct lfsck_tgt_descs *ltds,
2269                              struct lfsck_tgt_desc *ltd, __u16 type)
2270 {
2271         struct lfsck_component *com;
2272         int                     rc = 0;
2273         ENTRY;
2274
2275         LASSERT(lfsck->li_master);
2276
2277         spin_lock(&lfsck->li_lock);
2278         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_scan);
2279         if (com == NULL)
2280                 com = __lfsck_component_find(lfsck, type,
2281                                              &lfsck->li_list_double_scan);
2282         if (com != NULL)
2283                 lfsck_component_get(com);
2284         spin_unlock(&lfsck->li_lock);
2285
2286         if (com != NULL) {
2287                 struct lfsck_thread_info          *info  = lfsck_env_info(env);
2288                 struct lfsck_async_interpret_args *laia  = &info->lti_laia;
2289                 struct lfsck_request              *lr    = &info->lti_lr;
2290                 struct lfsck_assistant_data       *lad   = com->lc_data;
2291                 struct list_head                  *list;
2292                 struct list_head                  *phase_list;
2293                 struct ptlrpc_request_set         *set;
2294
2295                 set = ptlrpc_prep_set();
2296                 if (set == NULL) {
2297                         lfsck_component_put(env, com);
2298
2299                         RETURN(-ENOMEM);
2300                 }
2301
2302                 if (type == LFSCK_TYPE_LAYOUT) {
2303                         list = &ltd->ltd_layout_list;
2304                         phase_list = &ltd->ltd_layout_phase_list;
2305                 } else {
2306                         list = &ltd->ltd_namespace_list;
2307                         phase_list = &ltd->ltd_namespace_phase_list;
2308                 }
2309
2310                 spin_lock(&ltds->ltd_lock);
2311                 if (list_empty(list)) {
2312                         LASSERT(list_empty(phase_list));
2313                         spin_unlock(&ltds->ltd_lock);
2314                         ptlrpc_set_destroy(set);
2315
2316                         RETURN(0);
2317                 }
2318
2319                 list_del_init(phase_list);
2320                 list_del_init(list);
2321                 spin_unlock(&ltds->ltd_lock);
2322
2323                 memset(lr, 0, sizeof(*lr));
2324                 lr->lr_index = lfsck_dev_idx(lfsck);
2325                 lr->lr_event = LE_PEER_EXIT;
2326                 lr->lr_active = type;
2327                 lr->lr_status = LS_CO_PAUSED;
2328                 if (ltds == &lfsck->li_ost_descs)
2329                         lr->lr_flags = LEF_TO_OST;
2330
2331                 memset(laia, 0, sizeof(*laia));
2332                 laia->laia_com = com;
2333                 laia->laia_ltds = ltds;
2334                 atomic_inc(&ltd->ltd_ref);
2335                 laia->laia_ltd = ltd;
2336                 laia->laia_lr = lr;
2337
2338                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2339                                          lfsck_async_interpret_common,
2340                                          laia, LFSCK_NOTIFY);
2341                 if (rc != 0) {
2342                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for "
2343                                "co-stop for %s: rc = %d\n",
2344                                lfsck_lfsck2name(lfsck),
2345                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2346                                ltd->ltd_index, lad->lad_name, rc);
2347                         lfsck_tgt_put(ltd);
2348                 } else {
2349                         rc = ptlrpc_set_wait(set);
2350                 }
2351
2352                 ptlrpc_set_destroy(set);
2353                 lfsck_component_put(env, com);
2354         }
2355
2356         RETURN(rc);
2357 }
2358
2359 static int lfsck_async_interpret(const struct lu_env *env,
2360                                  struct ptlrpc_request *req,
2361                                  void *args, int rc)
2362 {
2363         struct lfsck_async_interpret_args *laia = args;
2364         struct lfsck_instance             *lfsck;
2365
2366         lfsck = container_of0(laia->laia_ltds, struct lfsck_instance,
2367                               li_mdt_descs);
2368         lfsck_interpret(env, lfsck, req, laia, rc);
2369         lfsck_tgt_put(laia->laia_ltd);
2370         if (rc != 0 && laia->laia_result != -EALREADY)
2371                 laia->laia_result = rc;
2372
2373         return 0;
2374 }
2375
2376 int lfsck_async_request(const struct lu_env *env, struct obd_export *exp,
2377                         struct lfsck_request *lr,
2378                         struct ptlrpc_request_set *set,
2379                         ptlrpc_interpterer_t interpreter,
2380                         void *args, int request)
2381 {
2382         struct lfsck_async_interpret_args *laia;
2383         struct ptlrpc_request             *req;
2384         struct lfsck_request              *tmp;
2385         struct req_format                 *format;
2386         int                                rc;
2387
2388         switch (request) {
2389         case LFSCK_NOTIFY:
2390                 format = &RQF_LFSCK_NOTIFY;
2391                 break;
2392         case LFSCK_QUERY:
2393                 format = &RQF_LFSCK_QUERY;
2394                 break;
2395         default:
2396                 CDEBUG(D_LFSCK, "%s: unknown async request %d: rc = %d\n",
2397                        exp->exp_obd->obd_name, request, -EINVAL);
2398                 return -EINVAL;
2399         }
2400
2401         req = ptlrpc_request_alloc(class_exp2cliimp(exp), format);
2402         if (req == NULL)
2403                 return -ENOMEM;
2404
2405         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, request);
2406         if (rc != 0) {
2407                 ptlrpc_request_free(req);
2408
2409                 return rc;
2410         }
2411
2412         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2413         *tmp = *lr;
2414         ptlrpc_request_set_replen(req);
2415
2416         laia = ptlrpc_req_async_args(req);
2417         *laia = *(struct lfsck_async_interpret_args *)args;
2418         if (laia->laia_com != NULL)
2419                 lfsck_component_get(laia->laia_com);
2420         req->rq_interpret_reply = interpreter;
2421         req->rq_allow_intr = 1;
2422         ptlrpc_set_add_req(set, req);
2423
2424         return 0;
2425 }
2426
2427 int lfsck_query_all(const struct lu_env *env, struct lfsck_component *com)
2428 {
2429         struct lfsck_thread_info          *info  = lfsck_env_info(env);
2430         struct lfsck_request              *lr    = &info->lti_lr;
2431         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
2432         struct lfsck_instance             *lfsck = com->lc_lfsck;
2433         struct lfsck_tgt_descs            *ltds  = &lfsck->li_mdt_descs;
2434         struct lfsck_tgt_desc             *ltd;
2435         struct ptlrpc_request_set         *set;
2436         int                                idx;
2437         int                                rc;
2438         ENTRY;
2439
2440         memset(lr, 0, sizeof(*lr));
2441         lr->lr_event = LE_QUERY;
2442         lr->lr_active = com->lc_type;
2443         lr->lr_flags = LEF_QUERY_ALL;
2444
2445         memset(laia, 0, sizeof(*laia));
2446         laia->laia_com = com;
2447         laia->laia_lr = lr;
2448
2449         set = ptlrpc_prep_set();
2450         if (set == NULL)
2451                 RETURN(-ENOMEM);
2452
2453 again:
2454         laia->laia_ltds = ltds;
2455         down_read(&ltds->ltd_rw_sem);
2456         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
2457                 ltd = lfsck_tgt_get(ltds, idx);
2458                 LASSERT(ltd != NULL);
2459
2460                 laia->laia_ltd = ltd;
2461                 up_read(&ltds->ltd_rw_sem);
2462                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2463                                          lfsck_async_interpret_common,
2464                                          laia, LFSCK_QUERY);
2465                 if (rc != 0) {
2466                         struct lfsck_assistant_data *lad = com->lc_data;
2467
2468                         CDEBUG(D_LFSCK, "%s: Fail to query %s %x for stat %s: "
2469                                "rc = %d\n", lfsck_lfsck2name(lfsck),
2470                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2471                                ltd->ltd_index, lad->lad_name, rc);
2472                         lfsck_reset_ltd_status(ltd, com->lc_type);
2473                         lfsck_tgt_put(ltd);
2474                 }
2475                 down_read(&ltds->ltd_rw_sem);
2476         }
2477         up_read(&ltds->ltd_rw_sem);
2478
2479         if (com->lc_type == LFSCK_TYPE_LAYOUT && !(lr->lr_flags & LEF_TO_OST)) {
2480                 ltds = &lfsck->li_ost_descs;
2481                 lr->lr_flags |= LEF_TO_OST;
2482                 goto again;
2483         }
2484
2485         rc = ptlrpc_set_wait(set);
2486         ptlrpc_set_destroy(set);
2487
2488         RETURN(rc);
2489 }
2490
2491 int lfsck_start_assistant(const struct lu_env *env, struct lfsck_component *com,
2492                           struct lfsck_start_param *lsp)
2493 {
2494         struct lfsck_instance           *lfsck   = com->lc_lfsck;
2495         struct lfsck_assistant_data     *lad     = com->lc_data;
2496         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
2497         struct ptlrpc_thread            *athread = &lad->lad_thread;
2498         struct lfsck_thread_args        *lta;
2499         struct task_struct              *task;
2500         int                              rc;
2501         ENTRY;
2502
2503         lad->lad_assistant_status = 0;
2504         lad->lad_post_result = 0;
2505         lad->lad_to_post = 0;
2506         lad->lad_to_double_scan = 0;
2507         lad->lad_in_double_scan = 0;
2508         lad->lad_exit = 0;
2509         lad->lad_advance_lock = false;
2510         thread_set_flags(athread, 0);
2511
2512         lta = lfsck_thread_args_init(lfsck, com, lsp);
2513         if (IS_ERR(lta))
2514                 RETURN(PTR_ERR(lta));
2515
2516         task = kthread_run(lfsck_assistant_engine, lta, lad->lad_name);
2517         if (IS_ERR(task)) {
2518                 rc = PTR_ERR(task);
2519                 CERROR("%s: cannot start LFSCK assistant thread for %s: "
2520                        "rc = %d\n", lfsck_lfsck2name(lfsck), lad->lad_name, rc);
2521                 lfsck_thread_args_fini(lta);
2522         } else {
2523                 struct l_wait_info lwi = { 0 };
2524
2525                 l_wait_event(mthread->t_ctl_waitq,
2526                              thread_is_running(athread) ||
2527                              thread_is_stopped(athread) ||
2528                              !thread_is_starting(mthread),
2529                              &lwi);
2530                 if (unlikely(!thread_is_starting(mthread)))
2531                         /* stopped by race */
2532                         rc = -ESRCH;
2533                 else if (unlikely(!thread_is_running(athread)))
2534                         rc = lad->lad_assistant_status;
2535                 else
2536                         rc = 0;
2537         }
2538
2539         RETURN(rc);
2540 }
2541
2542 int lfsck_checkpoint_generic(const struct lu_env *env,
2543                              struct lfsck_component *com)
2544 {
2545         struct lfsck_assistant_data     *lad     = com->lc_data;
2546         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2547         struct ptlrpc_thread            *athread = &lad->lad_thread;
2548         struct l_wait_info               lwi     = { 0 };
2549
2550         l_wait_event(mthread->t_ctl_waitq,
2551                      list_empty(&lad->lad_req_list) ||
2552                      !thread_is_running(mthread) ||
2553                      thread_is_stopped(athread),
2554                      &lwi);
2555
2556         if (!thread_is_running(mthread) || thread_is_stopped(athread))
2557                 return LFSCK_CHECKPOINT_SKIP;
2558
2559         return 0;
2560 }
2561
2562 void lfsck_post_generic(const struct lu_env *env,
2563                         struct lfsck_component *com, int *result)
2564 {
2565         struct lfsck_assistant_data     *lad     = com->lc_data;
2566         struct ptlrpc_thread            *athread = &lad->lad_thread;
2567         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2568         struct l_wait_info               lwi     = { 0 };
2569
2570         lad->lad_post_result = *result;
2571         if (*result <= 0)
2572                 lad->lad_exit = 1;
2573         lad->lad_to_post = 1;
2574
2575         CDEBUG(D_LFSCK, "%s: waiting for assistant to do %s post, rc = %d\n",
2576                lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, *result);
2577
2578         wake_up_all(&athread->t_ctl_waitq);
2579         l_wait_event(mthread->t_ctl_waitq,
2580                      (*result > 0 && list_empty(&lad->lad_req_list)) ||
2581                      thread_is_stopped(athread),
2582                      &lwi);
2583
2584         if (lad->lad_assistant_status < 0)
2585                 *result = lad->lad_assistant_status;
2586
2587         CDEBUG(D_LFSCK, "%s: the assistant has done %s post, rc = %d\n",
2588                lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, *result);
2589 }
2590
2591 int lfsck_double_scan_generic(const struct lu_env *env,
2592                               struct lfsck_component *com, int status)
2593 {
2594         struct lfsck_assistant_data     *lad     = com->lc_data;
2595         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2596         struct ptlrpc_thread            *athread = &lad->lad_thread;
2597         struct l_wait_info               lwi     = { 0 };
2598
2599         if (status != LS_SCANNING_PHASE2)
2600                 lad->lad_exit = 1;
2601         else
2602                 lad->lad_to_double_scan = 1;
2603
2604         CDEBUG(D_LFSCK, "%s: waiting for assistant to do %s double_scan, "
2605                "status %d\n",
2606                lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, status);
2607
2608         wake_up_all(&athread->t_ctl_waitq);
2609         l_wait_event(mthread->t_ctl_waitq,
2610                      lad->lad_in_double_scan ||
2611                      thread_is_stopped(athread),
2612                      &lwi);
2613
2614         CDEBUG(D_LFSCK, "%s: the assistant has done %s double_scan, "
2615                "status %d\n", lfsck_lfsck2name(com->lc_lfsck), lad->lad_name,
2616                lad->lad_assistant_status);
2617
2618         if (lad->lad_assistant_status < 0)
2619                 return lad->lad_assistant_status;
2620
2621         return 0;
2622 }
2623
2624 void lfsck_quit_generic(const struct lu_env *env,
2625                         struct lfsck_component *com)
2626 {
2627         struct lfsck_assistant_data     *lad     = com->lc_data;
2628         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2629         struct ptlrpc_thread            *athread = &lad->lad_thread;
2630         struct l_wait_info               lwi     = { 0 };
2631
2632         lad->lad_exit = 1;
2633         wake_up_all(&athread->t_ctl_waitq);
2634         l_wait_event(mthread->t_ctl_waitq,
2635                      thread_is_init(athread) ||
2636                      thread_is_stopped(athread),
2637                      &lwi);
2638 }
2639
2640 int lfsck_load_one_trace_file(const struct lu_env *env,
2641                               struct lfsck_component *com,
2642                               struct dt_object *parent,
2643                               struct dt_object **child,
2644                               const struct dt_index_features *ft,
2645                               const char *name, bool reset)
2646 {
2647         struct lfsck_instance *lfsck = com->lc_lfsck;
2648         struct dt_object *obj;
2649         int rc;
2650         ENTRY;
2651
2652         if (*child != NULL) {
2653                 struct dt_it *it;
2654                 const struct dt_it_ops *iops;
2655                 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid3;
2656
2657                 if (!reset)
2658                         RETURN(0);
2659
2660                 obj = *child;
2661                 rc = obj->do_ops->do_index_try(env, obj, ft);
2662                 if (rc)
2663                         /* unlink by force */
2664                         goto unlink;
2665
2666                 iops = &obj->do_index_ops->dio_it;
2667                 it = iops->init(env, obj, 0);
2668                 if (IS_ERR(it))
2669                         /* unlink by force */
2670                         goto unlink;
2671
2672                 fid_zero(fid);
2673                 rc = iops->get(env, it, (const struct dt_key *)fid);
2674                 if (rc >= 0) {
2675                         rc = iops->next(env, it);
2676                         iops->put(env, it);
2677                 }
2678                 iops->fini(env, it);
2679                 if (rc > 0)
2680                         /* "rc > 0" means the index file is empty. */
2681                         RETURN(0);
2682
2683 unlink:
2684                 /* The old index is not empty, remove it firstly. */
2685                 rc = local_object_unlink(env, lfsck->li_bottom, parent, name);
2686                 CDEBUG_LIMIT(rc ? D_ERROR : D_LFSCK,
2687                              "%s: unlink lfsck sub trace file %s: rc = %d\n",
2688                              lfsck_lfsck2name(com->lc_lfsck), name, rc);
2689                 if (rc)
2690                         RETURN(rc);
2691
2692                 if (*child) {
2693                         lfsck_object_put(env, *child);
2694                         *child = NULL;
2695                 }
2696         } else if (reset) {
2697                 goto unlink;
2698         }
2699
2700         obj = local_index_find_or_create(env, lfsck->li_los, parent, name,
2701                                          S_IFREG | S_IRUGO | S_IWUSR, ft);
2702         if (IS_ERR(obj))
2703                 RETURN(PTR_ERR(obj));
2704
2705         rc = obj->do_ops->do_index_try(env, obj, ft);
2706         if (rc)
2707                 lfsck_object_put(env, obj);
2708         else
2709                 *child = obj;
2710
2711         RETURN(rc);
2712 }
2713
2714 int lfsck_load_sub_trace_files(const struct lu_env *env,
2715                                struct lfsck_component *com,
2716                                const struct dt_index_features *ft,
2717                                const char *prefix, bool reset)
2718 {
2719         char *name = lfsck_env_info(env)->lti_key;
2720         struct lfsck_sub_trace_obj *lsto;
2721         int rc;
2722         int i;
2723
2724         for (i = 0, rc = 0, lsto = &com->lc_sub_trace_objs[0];
2725              i < LFSCK_STF_COUNT && rc == 0; i++, lsto++) {
2726                 snprintf(name, NAME_MAX, "%s_%02d", prefix, i);
2727                 rc = lfsck_load_one_trace_file(env, com,
2728                                 com->lc_lfsck->li_lfsck_dir,
2729                                 &lsto->lsto_obj, ft, name, reset);
2730         }
2731
2732         return rc;
2733 }
2734
2735 /* external interfaces */
2736
2737 int lfsck_get_speed(struct seq_file *m, struct dt_device *key)
2738 {
2739         struct lu_env           env;
2740         struct lfsck_instance  *lfsck;
2741         int                     rc;
2742         ENTRY;
2743
2744         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2745         if (rc != 0)
2746                 RETURN(rc);
2747
2748         lfsck = lfsck_instance_find(key, true, false);
2749         if (likely(lfsck != NULL)) {
2750                 seq_printf(m, "%u\n", lfsck->li_bookmark_ram.lb_speed_limit);
2751                 lfsck_instance_put(&env, lfsck);
2752         } else {
2753                 rc = -ENXIO;
2754         }
2755
2756         lu_env_fini(&env);
2757
2758         RETURN(rc);
2759 }
2760 EXPORT_SYMBOL(lfsck_get_speed);
2761
2762 int lfsck_set_speed(struct dt_device *key, __u32 val)
2763 {
2764         struct lu_env           env;
2765         struct lfsck_instance  *lfsck;
2766         int                     rc;
2767         ENTRY;
2768
2769         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2770         if (rc != 0)
2771                 RETURN(rc);
2772
2773         lfsck = lfsck_instance_find(key, true, false);
2774         if (likely(lfsck != NULL)) {
2775                 mutex_lock(&lfsck->li_mutex);
2776                 if (__lfsck_set_speed(lfsck, val))
2777                         rc = lfsck_bookmark_store(&env, lfsck);
2778                 mutex_unlock(&lfsck->li_mutex);
2779                 lfsck_instance_put(&env, lfsck);
2780         } else {
2781                 rc = -ENXIO;
2782         }
2783
2784         lu_env_fini(&env);
2785
2786         RETURN(rc);
2787 }
2788 EXPORT_SYMBOL(lfsck_set_speed);
2789
2790 int lfsck_get_windows(struct seq_file *m, struct dt_device *key)
2791 {
2792         struct lu_env           env;
2793         struct lfsck_instance  *lfsck;
2794         int                     rc;
2795         ENTRY;
2796
2797         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2798         if (rc != 0)
2799                 RETURN(rc);
2800
2801         lfsck = lfsck_instance_find(key, true, false);
2802         if (likely(lfsck != NULL)) {
2803                 seq_printf(m, "%u\n", lfsck->li_bookmark_ram.lb_async_windows);
2804                 lfsck_instance_put(&env, lfsck);
2805         } else {
2806                 rc = -ENXIO;
2807         }
2808
2809         lu_env_fini(&env);
2810
2811         RETURN(rc);
2812 }
2813 EXPORT_SYMBOL(lfsck_get_windows);
2814
2815 int lfsck_set_windows(struct dt_device *key, int val)
2816 {
2817         struct lu_env           env;
2818         struct lfsck_instance  *lfsck;
2819         int                     rc;
2820         ENTRY;
2821
2822         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2823         if (rc != 0)
2824                 RETURN(rc);
2825
2826         lfsck = lfsck_instance_find(key, true, false);
2827         if (likely(lfsck != NULL)) {
2828                 if (val < 1 || val > LFSCK_ASYNC_WIN_MAX) {
2829                         CWARN("%s: invalid async windows size that may "
2830                               "cause memory issues. The valid range is "
2831                               "[1 - %u].\n",
2832                               lfsck_lfsck2name(lfsck), LFSCK_ASYNC_WIN_MAX);
2833                         rc = -EINVAL;
2834                 } else if (lfsck->li_bookmark_ram.lb_async_windows != val) {
2835                         mutex_lock(&lfsck->li_mutex);
2836                         lfsck->li_bookmark_ram.lb_async_windows = val;
2837                         rc = lfsck_bookmark_store(&env, lfsck);
2838                         mutex_unlock(&lfsck->li_mutex);
2839                 }
2840                 lfsck_instance_put(&env, lfsck);
2841         } else {
2842                 rc = -ENXIO;
2843         }
2844
2845         lu_env_fini(&env);
2846
2847         RETURN(rc);
2848 }
2849 EXPORT_SYMBOL(lfsck_set_windows);
2850
2851 int lfsck_dump(struct seq_file *m, struct dt_device *key, enum lfsck_type type)
2852 {
2853         struct lu_env           env;
2854         struct lfsck_instance  *lfsck;
2855         struct lfsck_component *com;
2856         int                     rc;
2857         ENTRY;
2858
2859         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2860         if (rc != 0)
2861                 RETURN(rc);
2862
2863         lfsck = lfsck_instance_find(key, true, false);
2864         if (likely(lfsck != NULL)) {
2865                 com = lfsck_component_find(lfsck, type);
2866                 if (likely(com != NULL)) {
2867                         com->lc_ops->lfsck_dump(&env, com, m);
2868                         lfsck_component_put(&env, com);
2869                 } else {
2870                         rc = -ENOTSUPP;
2871                 }
2872
2873                 lfsck_instance_put(&env, lfsck);
2874         } else {
2875                 rc = -ENXIO;
2876         }
2877
2878         lu_env_fini(&env);
2879
2880         RETURN(rc);
2881 }
2882 EXPORT_SYMBOL(lfsck_dump);
2883
2884 static int lfsck_stop_all(const struct lu_env *env,
2885                           struct lfsck_instance *lfsck,
2886                           struct lfsck_stop *stop)
2887 {
2888         struct lfsck_thread_info          *info   = lfsck_env_info(env);
2889         struct lfsck_request              *lr     = &info->lti_lr;
2890         struct lfsck_async_interpret_args *laia   = &info->lti_laia;
2891         struct ptlrpc_request_set         *set;
2892         struct lfsck_tgt_descs            *ltds   = &lfsck->li_mdt_descs;
2893         struct lfsck_tgt_desc             *ltd;
2894         struct lfsck_bookmark             *bk     = &lfsck->li_bookmark_ram;
2895         __u32                              idx;
2896         int                                rc     = 0;
2897         int                                rc1    = 0;
2898         ENTRY;
2899
2900         LASSERT(stop->ls_flags & LPF_BROADCAST);
2901
2902         set = ptlrpc_prep_set();
2903         if (unlikely(set == NULL))
2904                 RETURN(-ENOMEM);
2905
2906         memset(lr, 0, sizeof(*lr));
2907         lr->lr_event = LE_STOP;
2908         lr->lr_index = lfsck_dev_idx(lfsck);
2909         lr->lr_status = stop->ls_status;
2910         lr->lr_version = bk->lb_version;
2911         lr->lr_active = LFSCK_TYPES_ALL;
2912         lr->lr_param = stop->ls_flags;
2913
2914         memset(laia, 0, sizeof(*laia));
2915         laia->laia_ltds = ltds;
2916         laia->laia_lr = lr;
2917         laia->laia_shared = 1;
2918
2919         down_read(&ltds->ltd_rw_sem);
2920         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
2921                 ltd = lfsck_tgt_get(ltds, idx);
2922                 LASSERT(ltd != NULL);
2923
2924                 laia->laia_ltd = ltd;
2925                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2926                                          lfsck_async_interpret, laia,
2927                                          LFSCK_NOTIFY);
2928                 if (rc != 0) {
2929                         lfsck_interpret(env, lfsck, NULL, laia, rc);
2930                         lfsck_tgt_put(ltd);
2931                         CERROR("%s: cannot notify MDT %x for LFSCK stop: "
2932                                "rc = %d\n", lfsck_lfsck2name(lfsck), idx, rc);
2933                         rc1 = rc;
2934                 }
2935         }
2936         up_read(&ltds->ltd_rw_sem);
2937
2938         rc = ptlrpc_set_wait(set);
2939         ptlrpc_set_destroy(set);
2940
2941         if (rc == 0)
2942                 rc = laia->laia_result;
2943
2944         if (rc == -EALREADY)
2945                 rc = 0;
2946
2947         if (rc != 0)
2948                 CERROR("%s: fail to stop LFSCK on some MDTs: rc = %d\n",
2949                        lfsck_lfsck2name(lfsck), rc);
2950
2951         RETURN(rc != 0 ? rc : rc1);
2952 }
2953
2954 static int lfsck_start_all(const struct lu_env *env,
2955                            struct lfsck_instance *lfsck,
2956                            struct lfsck_start *start)
2957 {
2958         struct lfsck_thread_info          *info   = lfsck_env_info(env);
2959         struct lfsck_request              *lr     = &info->lti_lr;
2960         struct lfsck_async_interpret_args *laia   = &info->lti_laia;
2961         struct ptlrpc_request_set         *set;
2962         struct lfsck_tgt_descs            *ltds   = &lfsck->li_mdt_descs;
2963         struct lfsck_tgt_desc             *ltd;
2964         struct lfsck_bookmark             *bk     = &lfsck->li_bookmark_ram;
2965         __u32                              idx;
2966         int                                rc     = 0;
2967         bool retry = false;
2968         ENTRY;
2969
2970         LASSERT(start->ls_flags & LPF_BROADCAST);
2971
2972         memset(lr, 0, sizeof(*lr));
2973         lr->lr_event = LE_START;
2974         lr->lr_index = lfsck_dev_idx(lfsck);
2975         lr->lr_speed = bk->lb_speed_limit;
2976         lr->lr_version = bk->lb_version;
2977         lr->lr_active = start->ls_active;
2978         lr->lr_param = start->ls_flags;
2979         lr->lr_async_windows = bk->lb_async_windows;
2980         lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN |
2981                        LSV_ASYNC_WINDOWS | LSV_CREATE_OSTOBJ |
2982                        LSV_CREATE_MDTOBJ;
2983
2984         memset(laia, 0, sizeof(*laia));
2985         laia->laia_ltds = ltds;
2986         laia->laia_lr = lr;
2987         laia->laia_shared = 1;
2988
2989 again:
2990         set = ptlrpc_prep_set();
2991         if (unlikely(!set))
2992                 RETURN(-ENOMEM);
2993
2994         down_read(&ltds->ltd_rw_sem);
2995         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
2996                 ltd = lfsck_tgt_get(ltds, idx);
2997                 LASSERT(ltd != NULL);
2998
2999                 if (retry && !ltd->ltd_retry_start) {
3000                         lfsck_tgt_put(ltd);
3001                         continue;
3002                 }
3003
3004                 laia->laia_ltd = ltd;
3005                 ltd->ltd_retry_start = 0;
3006                 ltd->ltd_layout_done = 0;
3007                 ltd->ltd_namespace_done = 0;
3008                 ltd->ltd_synced_failures = 0;
3009                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
3010                                          lfsck_async_interpret, laia,
3011                                          LFSCK_NOTIFY);
3012                 if (rc != 0) {
3013                         lfsck_interpret(env, lfsck, NULL, laia, rc);
3014                         lfsck_tgt_put(ltd);
3015                         CERROR("%s: cannot notify MDT %x for LFSCK "
3016                                "start, failout: rc = %d\n",
3017                                lfsck_lfsck2name(lfsck), idx, rc);
3018                         break;
3019                 }
3020         }
3021         up_read(&ltds->ltd_rw_sem);
3022
3023         if (rc != 0) {
3024                 ptlrpc_set_destroy(set);
3025
3026                 RETURN(rc);
3027         }
3028
3029         rc = ptlrpc_set_wait(set);
3030         ptlrpc_set_destroy(set);
3031
3032         if (rc == 0)
3033                 rc = laia->laia_result;
3034
3035         if (unlikely(rc == -EINPROGRESS)) {
3036                 retry = true;
3037                 set_current_state(TASK_INTERRUPTIBLE);
3038                 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC));
3039                 set_current_state(TASK_RUNNING);
3040                 if (!signal_pending(current) &&
3041                     thread_is_running(&lfsck->li_thread))
3042                         goto again;
3043
3044                 rc = -EINTR;
3045         }
3046
3047         if (rc != 0) {
3048                 struct lfsck_stop *stop = &info->lti_stop;
3049
3050                 CERROR("%s: cannot start LFSCK on some MDTs, "
3051                        "stop all: rc = %d\n",
3052                        lfsck_lfsck2name(lfsck), rc);
3053                 if (rc != -EALREADY) {
3054                         stop->ls_status = LS_FAILED;
3055                         stop->ls_flags = LPF_ALL_TGT | LPF_BROADCAST;
3056                         lfsck_stop_all(env, lfsck, stop);
3057                 }
3058         }
3059
3060         RETURN(rc);
3061 }
3062
3063 int lfsck_start(const struct lu_env *env, struct dt_device *key,
3064                 struct lfsck_start_param *lsp)
3065 {
3066         struct lfsck_start              *start  = lsp->lsp_start;
3067         struct lfsck_instance           *lfsck;
3068         struct lfsck_bookmark           *bk;
3069         struct ptlrpc_thread            *thread;
3070         struct lfsck_component          *com;
3071         struct l_wait_info               lwi    = { 0 };
3072         struct lfsck_thread_args        *lta;
3073         struct task_struct              *task;
3074         struct lfsck_tgt_descs          *ltds;
3075         struct lfsck_tgt_desc           *ltd;
3076         __u32                            idx;
3077         int                              rc     = 0;
3078         __u16                            valid  = 0;
3079         __u16                            flags  = 0;
3080         __u16                            type   = 1;
3081         ENTRY;
3082
3083         if (key->dd_rdonly)
3084                 RETURN(-EROFS);
3085
3086         lfsck = lfsck_instance_find(key, true, false);
3087         if (unlikely(lfsck == NULL))
3088                 RETURN(-ENXIO);
3089
3090         if (unlikely(lfsck->li_stopping))
3091                 GOTO(put, rc = -ENXIO);
3092
3093         /* System is not ready, try again later. */
3094         if (unlikely(lfsck->li_namespace == NULL ||
3095                      lfsck_dev_site(lfsck)->ss_server_fld == NULL))
3096                 GOTO(put, rc = -EINPROGRESS);
3097
3098         /* start == NULL means auto trigger paused LFSCK. */
3099         if (!start) {
3100                 if (list_empty(&lfsck->li_list_scan) ||
3101                     OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_AUTO))
3102                         GOTO(put, rc = 0);
3103         } else if (start->ls_flags & LPF_BROADCAST && !lfsck->li_master) {
3104                 CERROR("%s: only allow to specify '-A | -o' via MDS\n",
3105                        lfsck_lfsck2name(lfsck));
3106
3107                 GOTO(put, rc = -EPERM);
3108         }
3109
3110         bk = &lfsck->li_bookmark_ram;
3111         thread = &lfsck->li_thread;
3112         mutex_lock(&lfsck->li_mutex);
3113         spin_lock(&lfsck->li_lock);
3114         if (unlikely(thread_is_stopping(thread))) {
3115                 /* Someone is stopping the LFSCK. */
3116                 spin_unlock(&lfsck->li_lock);
3117                 GOTO(out, rc = -EBUSY);
3118         }
3119
3120         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
3121                 rc = -EALREADY;
3122                 if (unlikely(start == NULL)) {
3123                         spin_unlock(&lfsck->li_lock);
3124                         GOTO(out, rc);
3125                 }
3126
3127                 while (start->ls_active != 0) {
3128                         if (!(type & start->ls_active)) {
3129                                 type <<= 1;
3130                                 continue;
3131                         }
3132
3133                         com = __lfsck_component_find(lfsck, type,
3134                                                      &lfsck->li_list_scan);
3135                         if (com == NULL)
3136                                 com = __lfsck_component_find(lfsck, type,
3137                                                 &lfsck->li_list_double_scan);
3138                         if (com == NULL) {
3139                                 rc = -EOPNOTSUPP;
3140                                 break;
3141                         }
3142
3143                         if (com->lc_ops->lfsck_join != NULL) {
3144                                 rc = com->lc_ops->lfsck_join( env, com, lsp);
3145                                 if (rc != 0 && rc != -EALREADY)
3146                                         break;
3147                         }
3148                         start->ls_active &= ~type;
3149                         type <<= 1;
3150                 }
3151                 spin_unlock(&lfsck->li_lock);
3152                 GOTO(out, rc);
3153         }
3154         spin_unlock(&lfsck->li_lock);
3155
3156         lfsck->li_status = 0;
3157         lfsck->li_oit_over = 0;
3158         lfsck->li_start_unplug = 0;
3159         lfsck->li_drop_dryrun = 0;
3160         lfsck->li_new_scanned = 0;
3161
3162         /* For auto trigger. */
3163         if (start == NULL)
3164                 goto trigger;
3165
3166         start->ls_version = bk->lb_version;
3167
3168         if (start->ls_active != 0) {
3169                 struct lfsck_component *next;
3170
3171                 if (start->ls_active == LFSCK_TYPES_ALL)
3172                         start->ls_active = LFSCK_TYPES_SUPPORTED;
3173
3174                 if (start->ls_active & ~LFSCK_TYPES_SUPPORTED) {
3175                         start->ls_active &= ~LFSCK_TYPES_SUPPORTED;
3176                         GOTO(out, rc = -ENOTSUPP);
3177                 }
3178
3179                 list_for_each_entry_safe(com, next,
3180                                          &lfsck->li_list_scan, lc_link) {
3181                         if (!(com->lc_type & start->ls_active)) {
3182                                 rc = com->lc_ops->lfsck_post(env, com, 0,
3183                                                              false);
3184                                 if (rc != 0)
3185                                         GOTO(out, rc);
3186                         }
3187                 }
3188
3189                 while (start->ls_active != 0) {
3190                         if (type & start->ls_active) {
3191                                 com = __lfsck_component_find(lfsck, type,
3192                                                         &lfsck->li_list_idle);
3193                                 if (com != NULL)
3194                                         /* The component status will be updated
3195                                          * when its prep() is called later by
3196                                          * the LFSCK main engine. */
3197                                         list_move_tail(&com->lc_link,
3198                                                        &lfsck->li_list_scan);
3199                                 start->ls_active &= ~type;
3200                         }
3201                         type <<= 1;
3202                 }
3203         }
3204
3205         if (list_empty(&lfsck->li_list_scan)) {
3206                 /* The speed limit will be used to control both the LFSCK and
3207                  * low layer scrub (if applied), need to be handled firstly. */
3208                 if (start->ls_valid & LSV_SPEED_LIMIT) {
3209                         if (__lfsck_set_speed(lfsck, start->ls_speed_limit)) {
3210                                 rc = lfsck_bookmark_store(env, lfsck);
3211                                 if (rc != 0)
3212                                         GOTO(out, rc);
3213                         }
3214                 }
3215
3216                 goto trigger;
3217         }
3218
3219         if (start->ls_flags & LPF_RESET)
3220                 flags |= DOIF_RESET;
3221
3222         rc = lfsck_set_param(env, lfsck, start, !!(flags & DOIF_RESET));
3223         if (rc != 0)
3224                 GOTO(out, rc);
3225
3226         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
3227                 start->ls_active |= com->lc_type;
3228                 if (flags & DOIF_RESET) {
3229                         rc = com->lc_ops->lfsck_reset(env, com, false);
3230                         if (rc != 0)
3231                                 GOTO(out, rc);
3232                 }
3233         }
3234
3235         ltds = &lfsck->li_mdt_descs;
3236         down_read(&ltds->ltd_rw_sem);
3237         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
3238                 ltd = lfsck_ltd2tgt(ltds, idx);
3239                 LASSERT(ltd != NULL);
3240
3241                 ltd->ltd_layout_done = 0;
3242                 ltd->ltd_namespace_done = 0;
3243                 ltd->ltd_synced_failures = 0;
3244                 lfsck_reset_ltd_status(ltd, LFSCK_TYPE_NAMESPACE);
3245                 lfsck_reset_ltd_status(ltd, LFSCK_TYPE_LAYOUT);
3246                 list_del_init(&ltd->ltd_layout_phase_list);
3247                 list_del_init(&ltd->ltd_layout_list);
3248                 list_del_init(&ltd->ltd_namespace_phase_list);
3249                 list_del_init(&ltd->ltd_namespace_list);
3250         }
3251         up_read(&ltds->ltd_rw_sem);
3252
3253         ltds = &lfsck->li_ost_descs;
3254         down_read(&ltds->ltd_rw_sem);
3255         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
3256                 ltd = lfsck_ltd2tgt(ltds, idx);
3257                 LASSERT(ltd != NULL);
3258
3259                 ltd->ltd_layout_done = 0;
3260                 ltd->ltd_synced_failures = 0;
3261                 lfsck_reset_ltd_status(ltd, LFSCK_TYPE_LAYOUT);
3262                 list_del_init(&ltd->ltd_layout_phase_list);
3263                 list_del_init(&ltd->ltd_layout_list);
3264         }
3265         up_read(&ltds->ltd_rw_sem);
3266
3267 trigger:
3268         lfsck->li_args_dir = LUDA_64BITHASH | LUDA_VERIFY | LUDA_TYPE;
3269         if (bk->lb_param & LPF_DRYRUN)
3270                 lfsck->li_args_dir |= LUDA_VERIFY_DRYRUN;
3271
3272         if (start != NULL && start->ls_valid & LSV_ERROR_HANDLE) {
3273                 valid |= DOIV_ERROR_HANDLE;
3274                 if (start->ls_flags & LPF_FAILOUT)
3275                         flags |= DOIF_FAILOUT;
3276         }
3277
3278         if (start != NULL && start->ls_valid & LSV_DRYRUN) {
3279                 valid |= DOIV_DRYRUN;
3280                 if (start->ls_flags & LPF_DRYRUN)
3281                         flags |= DOIF_DRYRUN;
3282         }
3283
3284         if (!list_empty(&lfsck->li_list_scan))
3285                 flags |= DOIF_OUTUSED;
3286
3287         lfsck->li_args_oit = (flags << DT_OTABLE_IT_FLAGS_SHIFT) | valid;
3288         lta = lfsck_thread_args_init(lfsck, NULL, lsp);
3289         if (IS_ERR(lta))
3290                 GOTO(out, rc = PTR_ERR(lta));
3291
3292         __lfsck_set_speed(lfsck, bk->lb_speed_limit);
3293         spin_lock(&lfsck->li_lock);
3294         thread_set_flags(thread, SVC_STARTING);
3295         spin_unlock(&lfsck->li_lock);
3296         task = kthread_run(lfsck_master_engine, lta, "lfsck");
3297         if (IS_ERR(task)) {
3298                 rc = PTR_ERR(task);
3299                 CERROR("%s: cannot start LFSCK thread: rc = %d\n",
3300                        lfsck_lfsck2name(lfsck), rc);
3301                 lfsck_thread_args_fini(lta);
3302
3303                 GOTO(out, rc);
3304         }
3305
3306         l_wait_event(thread->t_ctl_waitq,
3307                      thread_is_running(thread) ||
3308                      thread_is_stopped(thread),
3309                      &lwi);
3310         if (start == NULL || !(start->ls_flags & LPF_BROADCAST)) {
3311                 lfsck->li_start_unplug = 1;
3312                 wake_up_all(&thread->t_ctl_waitq);
3313
3314                 GOTO(out, rc = 0);
3315         }
3316
3317         /* release lfsck::li_mutex to avoid deadlock. */
3318         mutex_unlock(&lfsck->li_mutex);
3319         rc = lfsck_start_all(env, lfsck, start);
3320         if (rc != 0) {
3321                 spin_lock(&lfsck->li_lock);
3322                 if (thread_is_stopped(thread)) {
3323                         spin_unlock(&lfsck->li_lock);
3324                 } else {
3325                         lfsck->li_status = LS_FAILED;
3326                         lfsck->li_flags = 0;
3327                         thread_set_flags(thread, SVC_STOPPING);
3328                         spin_unlock(&lfsck->li_lock);
3329
3330                         lfsck->li_start_unplug = 1;
3331                         wake_up_all(&thread->t_ctl_waitq);
3332                         l_wait_event(thread->t_ctl_waitq,
3333                                      thread_is_stopped(thread),
3334                                      &lwi);
3335                 }
3336         } else {
3337                 lfsck->li_start_unplug = 1;
3338                 wake_up_all(&thread->t_ctl_waitq);
3339         }
3340
3341         GOTO(put, rc);
3342
3343 out:
3344         mutex_unlock(&lfsck->li_mutex);
3345
3346 put:
3347         lfsck_instance_put(env, lfsck);
3348
3349         return rc < 0 ? rc : 0;
3350 }
3351 EXPORT_SYMBOL(lfsck_start);
3352
3353 int lfsck_stop(const struct lu_env *env, struct dt_device *key,
3354                struct lfsck_stop *stop)
3355 {
3356         struct lfsck_instance   *lfsck;
3357         struct ptlrpc_thread    *thread;
3358         struct l_wait_info       lwi    = { 0 };
3359         int                      rc     = 0;
3360         int                      rc1    = 0;
3361         ENTRY;
3362
3363         lfsck = lfsck_instance_find(key, true, false);
3364         if (unlikely(lfsck == NULL))
3365                 RETURN(-ENXIO);
3366
3367         thread = &lfsck->li_thread;
3368         if (stop && stop->ls_flags & LPF_BROADCAST && !lfsck->li_master) {
3369                 CERROR("%s: only allow to specify '-A' via MDS\n",
3370                        lfsck_lfsck2name(lfsck));
3371                 GOTO(put, rc = -EPERM);
3372         }
3373
3374         spin_lock(&lfsck->li_lock);
3375         /* The target is umounted */
3376         if (stop && stop->ls_status == LS_PAUSED)
3377                 lfsck->li_stopping = 1;
3378
3379         if (thread_is_init(thread) || thread_is_stopped(thread))
3380                 /* no error if LFSCK stopped already, or not started */
3381                 GOTO(unlock, rc = 0);
3382
3383         if (thread_is_stopping(thread))
3384                 /* Someone is stopping LFSCK. */
3385                 GOTO(unlock, rc = -EINPROGRESS);
3386
3387         if (stop) {
3388                 lfsck->li_status = stop->ls_status;
3389                 lfsck->li_flags = stop->ls_flags;
3390         } else {
3391                 lfsck->li_status = LS_STOPPED;
3392                 lfsck->li_flags = 0;
3393         }
3394
3395         thread_set_flags(thread, SVC_STOPPING);
3396
3397         if (lfsck->li_master) {
3398                 struct lfsck_component *com;
3399                 struct lfsck_assistant_data *lad;
3400
3401                 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
3402                         lad = com->lc_data;
3403                         spin_lock(&lad->lad_lock);
3404                         if (lad->lad_task != NULL)
3405                                 force_sig(SIGINT, lad->lad_task);
3406                         spin_unlock(&lad->lad_lock);
3407                 }
3408
3409                 list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
3410                         lad = com->lc_data;
3411                         spin_lock(&lad->lad_lock);
3412                         if (lad->lad_task != NULL)
3413                                 force_sig(SIGINT, lad->lad_task);
3414                         spin_unlock(&lad->lad_lock);
3415                 }
3416         }
3417
3418         wake_up_all(&thread->t_ctl_waitq);
3419         spin_unlock(&lfsck->li_lock);
3420         if (stop && stop->ls_flags & LPF_BROADCAST)
3421                 rc1 = lfsck_stop_all(env, lfsck, stop);
3422
3423         /* It was me set the status as 'stopping' just now, if it is not
3424          * 'stopping' now, then either stopped, or re-started by race. */
3425         l_wait_event(thread->t_ctl_waitq,
3426                      !thread_is_stopping(thread),
3427                      &lwi);
3428
3429         GOTO(put, rc = 0);
3430
3431 unlock:
3432         spin_unlock(&lfsck->li_lock);
3433 put:
3434         lfsck_instance_put(env, lfsck);
3435
3436         return rc != 0 ? rc : rc1;
3437 }
3438 EXPORT_SYMBOL(lfsck_stop);
3439
3440 int lfsck_in_notify_local(const struct lu_env *env, struct dt_device *key,
3441                           struct lfsck_req_local *lrl, struct thandle *th)
3442 {
3443         struct lfsck_instance *lfsck;
3444         struct lfsck_component *com;
3445         int rc = -EOPNOTSUPP;
3446         ENTRY;
3447
3448         lfsck = lfsck_instance_find(key, true, false);
3449         if (unlikely(!lfsck))
3450                 RETURN(-ENXIO);
3451
3452         com = lfsck_component_find(lfsck, lrl->lrl_active);
3453         if (likely(com && com->lc_ops->lfsck_in_notify_local)) {
3454                 rc = com->lc_ops->lfsck_in_notify_local(env, com, lrl, th);
3455                 lfsck_component_put(env, com);
3456         }
3457
3458         lfsck_instance_put(env, lfsck);
3459
3460         RETURN(rc);
3461 }
3462 EXPORT_SYMBOL(lfsck_in_notify_local);
3463
3464 int lfsck_in_notify(const struct lu_env *env, struct dt_device *key,
3465                     struct lfsck_request *lr)
3466 {
3467         int rc = -EOPNOTSUPP;
3468         ENTRY;
3469
3470         switch (lr->lr_event) {
3471         case LE_START: {
3472                 struct lfsck_start       *start = &lfsck_env_info(env)->lti_start;
3473                 struct lfsck_start_param  lsp;
3474
3475                 memset(start, 0, sizeof(*start));
3476                 start->ls_valid = lr->lr_valid;
3477                 start->ls_speed_limit = lr->lr_speed;
3478                 start->ls_version = lr->lr_version;
3479                 start->ls_active = lr->lr_active;
3480                 start->ls_flags = lr->lr_param & ~LPF_BROADCAST;
3481                 start->ls_async_windows = lr->lr_async_windows;
3482
3483                 lsp.lsp_start = start;
3484                 lsp.lsp_index = lr->lr_index;
3485                 lsp.lsp_index_valid = 1;
3486                 rc = lfsck_start(env, key, &lsp);
3487                 break;
3488         }
3489         case LE_STOP: {
3490                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
3491
3492                 memset(stop, 0, sizeof(*stop));
3493                 stop->ls_status = lr->lr_status;
3494                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
3495                 rc = lfsck_stop(env, key, stop);
3496                 break;
3497         }
3498         case LE_PHASE1_DONE:
3499         case LE_PHASE2_DONE:
3500         case LE_PEER_EXIT:
3501         case LE_CONDITIONAL_DESTROY:
3502         case LE_SET_LMV_MASTER:
3503         case LE_SET_LMV_SLAVE:
3504         case LE_PAIRS_VERIFY: {
3505                 struct lfsck_instance  *lfsck;
3506                 struct lfsck_component *com;
3507
3508                 lfsck = lfsck_instance_find(key, true, false);
3509                 if (unlikely(lfsck == NULL))
3510                         RETURN(-ENXIO);
3511
3512                 com = lfsck_component_find(lfsck, lr->lr_active);
3513                 if (likely(com)) {
3514                         rc = com->lc_ops->lfsck_in_notify(env, com, lr);
3515                         lfsck_component_put(env, com);
3516                 }
3517
3518                 lfsck_instance_put(env, lfsck);
3519                 break;
3520         }
3521         default:
3522                 break;
3523         }
3524
3525         RETURN(rc);
3526 }
3527 EXPORT_SYMBOL(lfsck_in_notify);
3528
3529 int lfsck_query(const struct lu_env *env, struct dt_device *key,
3530                 struct lfsck_request *req, struct lfsck_reply *rep,
3531                 struct lfsck_query *que)
3532 {
3533         struct lfsck_instance  *lfsck;
3534         struct lfsck_component *com;
3535         int                     i;
3536         int                     rc = 0;
3537         __u16                   type;
3538         ENTRY;
3539
3540         lfsck = lfsck_instance_find(key, true, false);
3541         if (unlikely(lfsck == NULL))
3542                 RETURN(-ENXIO);
3543
3544         if (que != NULL) {
3545                 if (que->lu_types == LFSCK_TYPES_ALL)
3546                         que->lu_types =
3547                                 LFSCK_TYPES_SUPPORTED & ~LFSCK_TYPE_SCRUB;
3548
3549                 if (que->lu_types & ~LFSCK_TYPES_SUPPORTED) {
3550                         que->lu_types &= ~LFSCK_TYPES_SUPPORTED;
3551
3552                         GOTO(out, rc = -ENOTSUPP);
3553                 }
3554
3555                 for (i = 0, type = 1 << i; i < LFSCK_TYPE_BITS;
3556                      i++, type = 1 << i) {
3557                         if (!(que->lu_types & type))
3558                                 continue;
3559
3560 again:
3561                         com = lfsck_component_find(lfsck, type);
3562                         if (unlikely(com == NULL))
3563                                 GOTO(out, rc = -ENOTSUPP);
3564
3565                         memset(que->lu_mdts_count[i], 0,
3566                                sizeof(__u32) * (LS_MAX + 1));
3567                         memset(que->lu_osts_count[i], 0,
3568                                sizeof(__u32) * (LS_MAX + 1));
3569                         que->lu_repaired[i] = 0;
3570                         rc = com->lc_ops->lfsck_query(env, com, req, rep,
3571                                                       que, i);
3572                         lfsck_component_put(env, com);
3573                         if  (rc < 0)
3574                                 GOTO(out, rc);
3575                 }
3576
3577                 if (!(que->lu_flags & LPF_WAIT))
3578                         GOTO(out, rc);
3579
3580                 for (i = 0, type = 1 << i; i < LFSCK_TYPE_BITS;
3581                      i++, type = 1 << i) {
3582                         if (!(que->lu_types & type))
3583                                 continue;
3584
3585                         if (que->lu_mdts_count[i][LS_SCANNING_PHASE1] != 0 ||
3586                             que->lu_mdts_count[i][LS_SCANNING_PHASE2] != 0 ||
3587                             que->lu_osts_count[i][LS_SCANNING_PHASE1] != 0 ||
3588                             que->lu_osts_count[i][LS_SCANNING_PHASE2] != 0) {
3589                                 struct l_wait_info lwi;
3590
3591                                 /* If it is required to wait, then sleep
3592                                  * 3 seconds and try to query again. */
3593                                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(3),
3594                                                        NULL,
3595                                                        LWI_ON_SIGNAL_NOOP,
3596                                                        NULL);
3597                                 rc = l_wait_event(lfsck->li_thread.t_ctl_waitq,
3598                                                   0, &lwi);
3599                                 if (rc == -ETIMEDOUT)
3600                                         goto again;
3601                         }
3602                 }
3603         } else {
3604                 com = lfsck_component_find(lfsck, req->lr_active);
3605                 if (likely(com != NULL)) {
3606                         rc = com->lc_ops->lfsck_query(env, com, req, rep,
3607                                                       que, -1);
3608                         lfsck_component_put(env, com);
3609                 } else {
3610                         rc = -ENOTSUPP;
3611                 }
3612         }
3613
3614         GOTO(out, rc);
3615
3616 out:
3617         lfsck_instance_put(env, lfsck);
3618         return rc;
3619 }
3620 EXPORT_SYMBOL(lfsck_query);
3621
3622 int lfsck_register_namespace(const struct lu_env *env, struct dt_device *key,
3623                              struct ldlm_namespace *ns)
3624 {
3625         struct lfsck_instance  *lfsck;
3626         int                     rc      = -ENXIO;
3627
3628         lfsck = lfsck_instance_find(key, true, false);
3629         if (likely(lfsck != NULL)) {
3630                 lfsck->li_namespace = ns;
3631                 lfsck_instance_put(env, lfsck);
3632                 rc = 0;
3633         }
3634
3635         return rc;
3636 }
3637 EXPORT_SYMBOL(lfsck_register_namespace);
3638
3639 int lfsck_register(const struct lu_env *env, struct dt_device *key,
3640                    struct dt_device *next, struct obd_device *obd,
3641                    lfsck_out_notify notify, void *notify_data, bool master)
3642 {
3643         struct lfsck_instance   *lfsck;
3644         struct dt_object        *root  = NULL;
3645         struct dt_object        *obj   = NULL;
3646         struct lu_fid           *fid   = &lfsck_env_info(env)->lti_fid;
3647         int                      rc;
3648         ENTRY;
3649
3650         lfsck = lfsck_instance_find(key, false, false);
3651         if (unlikely(lfsck != NULL))
3652                 RETURN(-EEXIST);
3653
3654         OBD_ALLOC_PTR(lfsck);
3655         if (lfsck == NULL)
3656                 RETURN(-ENOMEM);
3657
3658         mutex_init(&lfsck->li_mutex);
3659         spin_lock_init(&lfsck->li_lock);
3660         INIT_LIST_HEAD(&lfsck->li_link);
3661         INIT_LIST_HEAD(&lfsck->li_list_scan);
3662         INIT_LIST_HEAD(&lfsck->li_list_dir);
3663         INIT_LIST_HEAD(&lfsck->li_list_double_scan);
3664         INIT_LIST_HEAD(&lfsck->li_list_idle);
3665         INIT_LIST_HEAD(&lfsck->li_list_lmv);
3666         atomic_set(&lfsck->li_ref, 1);
3667         atomic_set(&lfsck->li_double_scan_count, 0);
3668         init_waitqueue_head(&lfsck->li_thread.t_ctl_waitq);
3669         lfsck->li_out_notify = notify;
3670         lfsck->li_out_notify_data = notify_data;
3671         lfsck->li_next = next;
3672         lfsck->li_bottom = key;
3673         lfsck->li_obd = obd;
3674
3675         rc = lfsck_tgt_descs_init(&lfsck->li_ost_descs);
3676         if (rc != 0)
3677                 GOTO(out, rc);
3678
3679         rc = lfsck_tgt_descs_init(&lfsck->li_mdt_descs);
3680         if (rc != 0)
3681                 GOTO(out, rc);
3682
3683         fid->f_seq = FID_SEQ_LOCAL_NAME;
3684         fid->f_oid = 1;
3685         fid->f_ver = 0;
3686         rc = local_oid_storage_init(env, key, fid, &lfsck->li_los);
3687         if (rc != 0)
3688                 GOTO(out, rc);
3689
3690         rc = dt_root_get(env, key, fid);
3691         if (rc != 0)
3692                 GOTO(out, rc);
3693
3694         root = dt_locate(env, key, fid);
3695         if (IS_ERR(root))
3696                 GOTO(out, rc = PTR_ERR(root));
3697
3698         if (unlikely(!dt_try_as_dir(env, root)))
3699                 GOTO(out, rc = -ENOTDIR);
3700
3701         lfsck->li_local_root_fid = *fid;
3702         if (master) {
3703                 lfsck->li_master = 1;
3704                 if (lfsck_dev_idx(lfsck) == 0) {
3705                         struct lu_fid *pfid = &lfsck_env_info(env)->lti_fid2;
3706                         const struct lu_name *cname;
3707
3708                         rc = dt_lookup(env, root,
3709                                 (struct dt_rec *)(&lfsck->li_global_root_fid),
3710                                 (const struct dt_key *)"ROOT");
3711                         if (rc != 0)
3712                                 GOTO(out, rc);
3713
3714                         obj = dt_locate(env, key, &lfsck->li_global_root_fid);
3715                         if (IS_ERR(obj))
3716                                 GOTO(out, rc = PTR_ERR(obj));
3717
3718                         if (unlikely(!dt_try_as_dir(env, obj)))
3719                                 GOTO(out, rc = -ENOTDIR);
3720
3721                         rc = dt_lookup(env, obj, (struct dt_rec *)fid,
3722                                 (const struct dt_key *)dotlustre);
3723                         if (rc != 0)
3724                                 GOTO(out, rc);
3725
3726                         lfsck_object_put(env, obj);
3727                         obj = dt_locate(env, key, fid);
3728                         if (IS_ERR(obj))
3729                                 GOTO(out, rc = PTR_ERR(obj));
3730
3731                         cname = lfsck_name_get_const(env, dotlustre,
3732                                                      strlen(dotlustre));
3733                         rc = lfsck_verify_linkea(env, obj, cname,
3734                                                  &lfsck->li_global_root_fid);
3735                         if (rc != 0)
3736                                 GOTO(out, rc);
3737
3738                         if (unlikely(!dt_try_as_dir(env, obj)))
3739                                 GOTO(out, rc = -ENOTDIR);
3740
3741                         *pfid = *fid;
3742                         rc = dt_lookup(env, obj, (struct dt_rec *)fid,
3743                                        (const struct dt_key *)lostfound);
3744                         if (rc != 0)
3745                                 GOTO(out, rc);
3746
3747                         lfsck_object_put(env, obj);
3748                         obj = dt_locate(env, key, fid);
3749                         if (IS_ERR(obj))
3750                                 GOTO(out, rc = PTR_ERR(obj));
3751
3752                         cname = lfsck_name_get_const(env, lostfound,
3753                                                      strlen(lostfound));
3754                         rc = lfsck_verify_linkea(env, obj, cname, pfid);
3755                         if (rc != 0)
3756                                 GOTO(out, rc);
3757
3758                         lfsck_object_put(env, obj);
3759                         obj = NULL;
3760                 }
3761         }
3762
3763         fid->f_seq = FID_SEQ_LOCAL_FILE;
3764         fid->f_oid = OTABLE_IT_OID;
3765         fid->f_ver = 0;
3766         obj = dt_locate(env, key, fid);
3767         if (IS_ERR(obj))
3768                 GOTO(out, rc = PTR_ERR(obj));
3769
3770         rc = obj->do_ops->do_index_try(env, obj, &dt_otable_features);
3771         if (rc != 0)
3772                 GOTO(out, rc);
3773
3774         lfsck->li_obj_oit = obj;
3775         obj = local_file_find_or_create(env, lfsck->li_los, root, LFSCK_DIR,
3776                                         S_IFDIR | S_IRUGO | S_IWUSR);
3777         if (IS_ERR(obj))
3778                 GOTO(out, rc = PTR_ERR(obj));
3779
3780         lu_object_get(&obj->do_lu);
3781         lfsck->li_lfsck_dir = obj;
3782         rc = lfsck_bookmark_setup(env, lfsck);
3783         if (rc != 0)
3784                 GOTO(out, rc);
3785
3786         if (master) {
3787                 rc = lfsck_fid_init(lfsck);
3788                 if (rc < 0)
3789                         GOTO(out, rc);
3790
3791                 rc = lfsck_namespace_setup(env, lfsck);
3792                 if (rc < 0)
3793                         GOTO(out, rc);
3794         }
3795
3796         rc = lfsck_layout_setup(env, lfsck);
3797         if (rc < 0)
3798                 GOTO(out, rc);
3799
3800         /* XXX: more LFSCK components initialization to be added here. */
3801
3802         rc = lfsck_instance_add(lfsck);
3803         if (rc == 0)
3804                 rc = lfsck_add_target_from_orphan(env, lfsck);
3805 out:
3806         if (obj != NULL && !IS_ERR(obj))
3807                 lfsck_object_put(env, obj);
3808         if (root != NULL && !IS_ERR(root))
3809                 lfsck_object_put(env, root);
3810         if (rc != 0)
3811                 lfsck_instance_cleanup(env, lfsck);
3812         return rc;
3813 }
3814 EXPORT_SYMBOL(lfsck_register);
3815
3816 void lfsck_degister(const struct lu_env *env, struct dt_device *key)
3817 {
3818         struct lfsck_instance *lfsck;
3819
3820         lfsck = lfsck_instance_find(key, false, true);
3821         if (lfsck != NULL)
3822                 lfsck_instance_put(env, lfsck);
3823 }
3824 EXPORT_SYMBOL(lfsck_degister);
3825
3826 int lfsck_add_target(const struct lu_env *env, struct dt_device *key,
3827                      struct dt_device *tgt, struct obd_export *exp,
3828                      __u32 index, bool for_ost)
3829 {
3830         struct lfsck_instance   *lfsck;
3831         struct lfsck_tgt_desc   *ltd;
3832         int                      rc;
3833         ENTRY;
3834
3835         OBD_ALLOC_PTR(ltd);
3836         if (ltd == NULL)
3837                 RETURN(-ENOMEM);
3838
3839         ltd->ltd_tgt = tgt;
3840         ltd->ltd_key = key;
3841         ltd->ltd_exp = exp;
3842         INIT_LIST_HEAD(&ltd->ltd_orphan_list);
3843         INIT_LIST_HEAD(&ltd->ltd_layout_list);
3844         INIT_LIST_HEAD(&ltd->ltd_layout_phase_list);
3845         INIT_LIST_HEAD(&ltd->ltd_namespace_list);
3846         INIT_LIST_HEAD(&ltd->ltd_namespace_phase_list);
3847         atomic_set(&ltd->ltd_ref, 1);
3848         ltd->ltd_index = index;
3849
3850         spin_lock(&lfsck_instance_lock);
3851         lfsck = __lfsck_instance_find(key, true, false);
3852         if (lfsck == NULL) {
3853                 if (for_ost)
3854                         list_add_tail(&ltd->ltd_orphan_list,
3855                                       &lfsck_ost_orphan_list);
3856                 else
3857                         list_add_tail(&ltd->ltd_orphan_list,
3858                                       &lfsck_mdt_orphan_list);
3859                 spin_unlock(&lfsck_instance_lock);
3860
3861                 RETURN(0);
3862         }
3863         spin_unlock(&lfsck_instance_lock);
3864
3865         rc = __lfsck_add_target(env, lfsck, ltd, for_ost, false);
3866         if (rc != 0)
3867                 lfsck_tgt_put(ltd);
3868
3869         lfsck_instance_put(env, lfsck);
3870
3871         RETURN(rc);
3872 }
3873 EXPORT_SYMBOL(lfsck_add_target);
3874
3875 void lfsck_del_target(const struct lu_env *env, struct dt_device *key,
3876                       struct dt_device *tgt, __u32 index, bool for_ost)
3877 {
3878         struct lfsck_instance   *lfsck;
3879         struct lfsck_tgt_descs  *ltds;
3880         struct lfsck_tgt_desc   *ltd;
3881         struct list_head        *head;
3882
3883         if (for_ost)
3884                 head = &lfsck_ost_orphan_list;
3885         else
3886                 head = &lfsck_mdt_orphan_list;
3887
3888         spin_lock(&lfsck_instance_lock);
3889         list_for_each_entry(ltd, head, ltd_orphan_list) {
3890                 if (ltd->ltd_tgt == tgt) {
3891                         list_del_init(&ltd->ltd_orphan_list);
3892                         spin_unlock(&lfsck_instance_lock);
3893                         lfsck_tgt_put(ltd);
3894
3895                         return;
3896                 }
3897         }
3898
3899         ltd = NULL;
3900         lfsck = __lfsck_instance_find(key, true, false);
3901         spin_unlock(&lfsck_instance_lock);
3902         if (unlikely(lfsck == NULL))
3903                 return;
3904
3905         if (for_ost)
3906                 ltds = &lfsck->li_ost_descs;
3907         else
3908                 ltds = &lfsck->li_mdt_descs;
3909
3910         down_write(&ltds->ltd_rw_sem);
3911         LASSERT(ltds->ltd_tgts_bitmap != NULL);
3912
3913         if (unlikely(index >= ltds->ltd_tgts_bitmap->size))
3914                 goto unlock;
3915
3916         ltd = lfsck_ltd2tgt(ltds, index);
3917         if (unlikely(ltd == NULL))
3918                 goto unlock;
3919
3920         LASSERT(ltds->ltd_tgtnr > 0);
3921
3922         ltds->ltd_tgtnr--;
3923         cfs_bitmap_clear(ltds->ltd_tgts_bitmap, index);
3924         lfsck_assign_tgt(ltds, NULL, index);
3925
3926 unlock:
3927         if (ltd == NULL) {
3928                 if (for_ost)
3929                         head = &lfsck->li_ost_descs.ltd_orphan;
3930                 else
3931                         head = &lfsck->li_mdt_descs.ltd_orphan;
3932
3933                 list_for_each_entry(ltd, head, ltd_orphan_list) {
3934                         if (ltd->ltd_tgt == tgt) {
3935                                 list_del_init(&ltd->ltd_orphan_list);
3936                                 break;
3937                         }
3938                 }
3939         }
3940
3941         up_write(&ltds->ltd_rw_sem);
3942         if (ltd != NULL) {
3943                 spin_lock(&ltds->ltd_lock);
3944                 ltd->ltd_dead = 1;
3945                 spin_unlock(&ltds->ltd_lock);
3946                 lfsck_stop_notify(env, lfsck, ltds, ltd, LFSCK_TYPE_NAMESPACE);
3947                 lfsck_stop_notify(env, lfsck, ltds, ltd, LFSCK_TYPE_LAYOUT);
3948                 lfsck_tgt_put(ltd);
3949         }
3950
3951         lfsck_instance_put(env, lfsck);
3952 }
3953 EXPORT_SYMBOL(lfsck_del_target);
3954
3955 static int __init lfsck_init(void)
3956 {
3957         int rc;
3958
3959         INIT_LIST_HEAD(&lfsck_instance_list);
3960         INIT_LIST_HEAD(&lfsck_ost_orphan_list);
3961         INIT_LIST_HEAD(&lfsck_mdt_orphan_list);
3962         lfsck_key_init_generic(&lfsck_thread_key, NULL);
3963         rc = lu_context_key_register(&lfsck_thread_key);
3964         if (!rc) {
3965                 tgt_register_lfsck_in_notify_local(lfsck_in_notify_local);
3966                 tgt_register_lfsck_in_notify(lfsck_in_notify);
3967                 tgt_register_lfsck_query(lfsck_query);
3968         }
3969
3970         return rc;
3971 }
3972
3973 static void __exit lfsck_exit(void)
3974 {
3975         struct lfsck_tgt_desc *ltd;
3976         struct lfsck_tgt_desc *next;
3977
3978         LASSERT(list_empty(&lfsck_instance_list));
3979
3980         list_for_each_entry_safe(ltd, next, &lfsck_ost_orphan_list,
3981                                  ltd_orphan_list) {
3982                 list_del_init(&ltd->ltd_orphan_list);
3983                 lfsck_tgt_put(ltd);
3984         }
3985
3986         list_for_each_entry_safe(ltd, next, &lfsck_mdt_orphan_list,
3987                                  ltd_orphan_list) {
3988                 list_del_init(&ltd->ltd_orphan_list);
3989                 lfsck_tgt_put(ltd);
3990         }
3991
3992         lu_context_key_degister(&lfsck_thread_key);
3993 }
3994
3995 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3996 MODULE_DESCRIPTION("Lustre File System Checker");
3997 MODULE_VERSION(LUSTRE_VERSION_STRING);
3998 MODULE_LICENSE("GPL");
3999
4000 module_init(lfsck_init);
4001 module_exit(lfsck_exit);